gptme changelog

These are the release notes for gptme version v0.36.0.

Contributors

Thanks to everyone who contributed to this release:

@daeho-ro, @Ehsan-hash, @erikbjare, @March-7, @original4422, @sarankumar1325, @simpleqt, @TimeToBuildBob, @wozhenbang2004

Changelog

Changes since 0c131bb0d642fcdd1edb2c6fec02eccfe7f71a07:

📦 gptme

✨ Features (19)

• feat(providers): add OpenCode Go DeepSeek support (#4516) (46a43eb40)
• feat(tauri): Windows Authenticode signing for installer and sidecar (#4510) (93bf79566)
• feat(complete): discover opt-in project verification (#3755) (d5d935c34)
• feat(models): update gptme managed default to openrouter/anthropic/claude-sonnet-5.5 (#4500) (33f2c9cb4)
• feat(review): preserve multi-pass consensus evidence on ReviewFinding (#4442) (d4204c3a4)
• feat(review): add per-finding evidence fields and artifact identity fields (#4441) (4c41b85f0)
• feat: support Python 3.15 (#4410) (a7602d6de)
• feat(tools): policy-block helper — consistent marker + finality text at all block sites (#3959) (945c12d7d)
• feat(hooks): inject working-tree diff at session start when dirty (#4020) (07c309e39)
• feat(review): retain full rejected_findings objects in ReviewArtifact (#4406) (2098420a9)
• feat(review): extend ReviewArtifact with production evidence fields (#4405) (1d03bcdaf)
• feat(review): add JSON Schema + validator for ReviewArtifact v2 (slice 2d) (#4404) (787180fea)
• feat(python): auto-activate kernel backend when a project environment is detected (#4393) (7ec4b5e1b)
• feat(webui): add group-turns toggle and fix tool-result indent edge cases (#4376) (#4386) (9218f8c24)
• feat(review): extend ReviewArtifact with union schema fields (slice 2a) (#4390) (01308d1d6)
• feat(python): use 'uv run --with ipykernel' when project env lacks ipykernel (#4379) (4b856f921)
• feat(review): ingest ai-review-local artifacts + extend finding schema (#4388) (880e22442)
• feat(webui): indent tool results under their step in expanded turn groups (#4380) (10a23b6de)
• feat(tui): show an agent turn as one block, tool results under their step (#4375) (0f49283ae)

🐛 Fixes (229)

Click to expand

• fix(tui): preserve queued prompt boundaries after interrupt (#4758) (b7ca28976)
• fix: distinguish retained log size from provider context usage (#4757) (90e1d1ec1)
• fix(gh): return early from pr checks on failure and bound waits (#4755) (4432f1020)
• fix(setup): recognize AQ. Gemini API keys (#4754) (67f35d8ad)
• fix: configurable summary models and refresh Gemini defaults (#4749) (cb2bf23c0)
• fix(export): preserve literal transcript examples when filtering (#4671) (151278283)
• fix(convert): reject special-file output destinations (#4598) (b66fff581)
• fix(browser): keep live auth state across open_page navigation (#4726) (36c97a710)
• fix(convert): preserve percent-pattern image source identity (#4676) (1deef389b)
• fix(chats): keep cleanup dry-run free of lifecycle writes (#4744) (1c3dbd063)
• fix(browser): validate state before discarding active session (#4646) (83ff72528)
• fix(webui): stabilize settings updater for effect consumers (#4581) (8a5350849)
• fix(patch_many): preserve literal headers through confirmation (#4578) (14bc506e4)
• fix(patch_many): stage writes and restore failed publications (#4564) (44f186f34

…

These are the release notes for gptme version v0.35.1.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare, @TimeToBuildBob

Changelog

Changes since v0.35.0:

📦 gptme

✨ Features (1)

• feat(python): run IPython in a separate Jupyter kernel process (#4359) (7dc2d7c87)

🐛 Fixes (4)

• fix(shell): shellcheck blocks error-severity findings, hides style advice (#4374) (c2d437ea1)
• fix(tui): no blinks or jumps while a turn runs; frame-level regression tests (#4372) (4ed122a37)
• fix(release): pin concrete Python for uv build in publish job (#4370) (c72b136ce)
• fix(llm): avoid rebuilding grok client per-request; reuse HTTP session in openai-subscription (#4365) (88bc08d6b)

🔨 Misc (1)

• chore: bump version to 0.35.1 (v0.35.1)

Full Changelog: https://github.com/gptme/gptme/compare/v0.35.0...v0.35.1

These are the release notes for gptme version v0.35.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare, @harimaruthachalam, @jaideeppyne, @Thibaultjaigu, @TimeToBuildBob, @vanshtaneja23

Changelog

Changes since v0.34.0:

📦 gptme

✨ Features (61)

• feat(shell): memory guard kills the agent's largest command when system memory runs low (#4348) (6aa340f35)
• feat(tui): queued prompts steer the running turn at its next step (#4346) (e4fb3d976)
• feat(tui,ipython): Ctrl+C stops a running tool in the TUI; ipython cells time out (#4354) (d8a54a974)
• feat(tui): show the running tool call, its elapsed time and timeout (#4344) (804273354)
• feat(lessons): capture private final-item evidence for precision readouts (#4340) (ffd45727c)
• feat(tui): command-driven status line ([tui] statusline) (#4339) (3160b45f1)
• feat(context): add max_tokens to ContextSelectorConfig; use it in LLMSelector (#4336) (c4f1e463d)
• feat(lessons): record final native context emissions (#4328) (47673511b)
• feat(compaction): record 'chunked' and 'native' methods in compaction.jsonl (#4318) (7413443fc)
• feat(compaction): summarize overflowed history in fresh-context chunks (#4309) (fe24c6c0c)
• feat(lessons): add optional GPTME_LESSONS_MAX_PER_TURN limit (#4294) (7835c616f)
• feat(server): apply model default_tool_format when webui omits tool_format (#4245) (3e9169958)
• feat(compaction): capability-gated Anthropic native compaction with signed-block replay (Phase 2) (#4244) (6fe8ad4e1)
• feat(lessons): add opt-in shared selector command adapter (#4283) (6b4338cfc)
• feat(site): add gptme.org/stats/ page from gptme/stats (#4279) (381c29320)
• feat(config): persist --agent-profile and re-apply it on resume (#4040) (9511db3b7)
• feat(eval): add constraint-satisfaction eval suite (#4082) (4a9eb56ad)
• feat(server): share auto-generated token with local clients via token file (#4061) (6fdbc7bac)
• feat(compaction): run checkpoint through normal tool loop (#4223) (56ae66070)
• feat(models): register Mistral Large 4 (#4229) (83401e8bf)
• feat(compaction): remind normal tool loop before context budget (#4174) (021e57777)
• feat(webui): expose server-owned global MCP settings (#4218) (945cc4da9)
• feat(server): structured GET/PUT /api/v2/user/config/mcp for global MCP config (#4095) (b8c8527c9)
• feat(util): add gptme-util sound to play notification sounds from other harnesses (#4205) (70dab56ff)
• feat(hooks): behavioral anomaly watchdog — scope_escape, write_storm, novel_host (#3953) (3c8408463)
• feat(subagent): context_mode=fork, per-call reasoning_effort, usage reporting (#4096) (101b8a7e4)
• feat(compaction): Phase 2 — compact_instructions, keep_recent, structured checkpoint (#3812) (#3966) (e632523b5)
• feat(watch): refuse sleep-N polling chains in the shell while watch is loaded (#4094) (b3eacfcab)
• feat(llm): client-side degeneration guard for OpenRouter streams (#4034) (9f1697c21)
• feat(webui): thread sseToken through to ApiClient EventSource ?token= (#4035) (7f172f7fe)
• feat(eval): add OpenShellExecutionEnv with kernel-enforced policy isolation (#4027) (e549e7cf6)
• feat(llm): attribute OpenRouter serving provider from response body (#4031) (43b17c69d)
• feat(llm): record provider-served model in message metadata (#4026

…

These are the release notes for gptme version v0.34.0.

Contributors

Thanks to everyone who contributed to this release:

@AmaLS367, @aniruddhaadak80, @erikbjare, @simpleqt, @TimeToBuildBob

Changelog

Changes since v0.33.0:

📦 gptme

✨ Features (61)

• feat(webui): add gptme-host:seed-prompt to pre-fill chat input from host (#3870) (437eca576)
• feat(memory): add search and export commands to complete #3734 CLI table (#3869) (a0418fe30)
• feat(memory): migrate-knowledge-jsonl command + deprecate knowledge CLI (#3863) (19f3ece8b)
• feat(cloud): default service URL → auth.gptme.ai custom domain (#3866) (d78c71188)
• feat(knowledge): wire search to gptme-rag semantic ranking (#3865) (73074e58d)
• feat(knowledge): generalize schema with entry_type (problem_resolution, decision, fact, how_to, note) (#3864) (b42d24b85)
• feat(config): expose preview guidance through runtime prompt fragments (#3821) (c74602087)
• feat(site): deploy design-v3 landing page to gptme.org (#3849) (170f28bb4)
• feat(site): prototype of new gptme.org site (design v3) (#3839) (0881c1d98)
• feat(memory): trigger keyworded entries through the lesson matcher (#3817) (1ced9ccca)
• feat(subagent): persist registry metadata and rehydrate after restart (#3834) (085bd7bec)
• feat(skills): persist CostTracker ownership across TUI/server workers (#3833) (3c4326778)
• feat(subagent): continue completed child conversations (#3820) (e4e0f6980)
• feat(skills): export lifecycle metrics and measured cost windows (#3793) (6f9a5bdfc)
• feat(tools): conditional tool docs + companion tools (requires_tools) (#3801) (925eab097)
• feat(memory): use layered MemoryStore in workspace prompt (#3789) (31a385390)
• feat(memory): preserve writer provenance through CLI metadata (#3792) (1ecb322fc)
• feat(memory): persist always-on selection and byte budget (#3791) (587048f44)
• feat(models): add deepseek/deepseek-v4.1-flash (#3799) (0241c2592)
• feat(skills): record server and TUI invocation outcomes (#3787) (0e7033754)
• feat(memory): Codex/AGENTS.md integration for cross-harness memory (#3764) (b5374121c)
• feat(models): generate recommended-model docs from code; add DeepSeek V4 Flash 0731, GLM 5.3 Flash, GPT-6 Astra, Grok 4.6; OpenRouter provider allowlists (#3786) (a35b7fdf5)
• feat(cli): add --track-tokens flag to print running token count per LLM call (#3646) (235a3fb90)
• feat(skills): persist explicit invocation lifecycle evidence (#3750) (fc8024909)
• feat(llm): make reasoning effort controllable and observable for OpenAI-compatible providers (#3781) (75c65631c)
• feat(config): trust-on-first-use gate for project-level shell execution (#3782) (dbd5f6173)
• feat(hooks): add guardrail TOOL_CONFIRM hook for headless safety (#3598) (#3765) (c176aa62d)
• feat(server): add /preview/{port}/ proxy for cloud port exposure (gptme-cloud#910) (#3767) (5e6083c91)
• feat(memory): add cross-harness lexical recall (#3740) (d4bcf8658)
• feat(tools): add ToolSpec.read_only flag; auto-approve in cli_confirm_hook (#3733) (9573ef16e)
• feat(memory): gptme.memory package + gptme-util memory CLI (layered roots, CC-compatible) (#3735) (21334d854)
• feat(shell): stream one-shot command output while it runs (#3579) (e3d15c571)
• feat(memory): add memory write tool for cross-runtime memory sharing (#3638) (39810873a)
• feat(dataset): trajectory-to-env pipeline for fine-tuning dataset construction (#3719) (86fc95533)
• feat(cli): add gptme capabilities export (#3705) (6b7d53d1a)
• feat(harness): actuate request_tool_change — enable/disable tools mid-session (Phase 2) (#3656) (d3b39582d)
• feat(lessons): class-aware dropout for validated_core (Phase 1) (#3700

…

These are the release notes for gptme version v0.33.0.

Contributors

Thanks to everyone who contributed to this release:

@AlexFerroni02, @eeshsaxena, @erikbjare, @illiand, @sujeito-operator, @TimeToBuildBob

Changelog

Changes since v0.32.1:

📦 gptme

✨ Features (76)

• feat(autocompact): add keep_head to protect task context from compaction (#3567) (ab731347b)
• feat(prompts): Add code editing strategy guidance for improved model performance (#3530) (8fb767452)
• feat(cli): fatal error envelope + exit taxonomy for non-interactive mode (#3568) (3fb862dcc)
• feat(prompts): end responses with a concrete next step (#3559) (a93bf261f)
• feat(tools): add read-only audit preset (#3543) (7adb7f765)
• feat(cli): add gptme explain for offline concept answers (#3542) (bb48ab5be)
• feat(cli): add --version-short output (#3536) (bff711a52)
• feat(models): refresh MiniMax plugin registry (#3533) (eb0dae860)
• feat(cli): add JSON output to status command (#3528) (f232004be)
• feat(tools/hashline_edit): 3-way merge recovery when file changes between read and edit (#3476 Phase 2) (#3520) (c3bdfb46d)
• feat(attestation): Phase 1 registry bridge - wire model-capability-registry (#3525) (fce2fbe69)
• feat(tools/hashline_edit): add register operations (CUT @name / PUT @name) (#3497) (cbce4062e)
• feat(review): make reviewer toolset configurable with a read-only preset (#3512) (6b3fde6fc)
• feat(lessons): Stage 1 shadow logging — policy_class/policy_version in dropout logs (#3455) (bd99a0944)
• feat(review): add gptme-util review pr subcommand — complete the unified review pipeline (gptme#3442) (#3462) (374f94304)
• feat(llm/webui): capture and display resolved OpenRouter subprovider in message metadata (#3486) (34e8848a8)
• feat(tools/hashline_edit): add PUT N*: block-aware replace (#3476 Phase 1 remainder) (#3480) (db8fe3881)
• feat(webui): add trajectory copy commands (#3491) (7c29b4177)
• feat(webui): add copy trajectory as markdown with detail level toggles (#3485) (786f78642)
• feat(review): add --trusted-reviewer guard for artifact mode (#3470) (702d9c219)
• feat(tools): hashline-inspired snapshot-anchored edit format (#3477) (8fcff6ae8)
• feat(webui/model-picker): add free-text custom model ID entry (#3466) (f5c6c8be5)
• feat(review): artifact mode + structured handoff for review-watch (#3442) (#3449) (9130a018c)
• feat(cli): add global stats for CLI subcommand for usage & cost analytics (#3400) (8d88b862e)
• feat(review): converge review-watch and shared review pipeline (#3442) (#3443) (4a21419d6)
• feat(timings): persist per-step timing breakdown in session records (#3436) (b0adaa7c7)
• feat(util): add review-watch command for autonomous PR review iteration (#3437) (81342c4e6)
• feat(browser): support custom executable path in GPTME_BROWSER_ENGINE (#3439) (08d61d970)
• feat(cli): add 'gptme providers add' command for interactive custom provider setup (#3438) (542f64035)
• feat(android): sign APK/AAB and verify signature before release upload (#3409) (55ac56712)
• feat(cost): add session budget warnings (#3423) (be1e0b4f8)
• feat(attestation): persist ModelSelectionTrace to logdir and embed in attestations (#3424) (5721cd0a7)
• feat(site): add built-with-gptme badge (#3410) (3e146fcce)
• feat(attestation): add ModelSelectionTrace for durable model provenance (Phase 0) (#3416) (1b4616ee0)
• feat(webui): add Tool Activity Panel to right sidebar (#3413) (a598f55d2)
• feat(tutorial): add gptme-tutorial interactive onboarding mode (#3420) (c284f67e7)

…

These are the release notes for gptme version v0.32.1.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare, @TimeToBuildBob

Changelog

Changes since v0.32.0:

📦 gptme

✨ Features (27)

• feat(webui): warn on server API incompatibility (#3271) (6c5c979b7)
• feat(context): expose initial prompt to context_cmd (#3270) (47c30c14c)
• feat(webui): add BranchMapPanel to right sidebar (#3264) (ab4ae0950)
• feat(subagent): cooperative cancel checkpoint — make cancel actually stop thread-mode subagents (#3261) (9acd22a61)
• feat(subagent): stop thread agents at control checkpoints (#3262) (808c1af76)
• feat(subagent): add max_concurrent fleet cap to subagent_parallel() (#3222) (6eda8b7d4)
• feat(anti-slop): add LLM-slop detector and quality gate CLI (#3245) (55bc30398)
• feat(llm): stream idle-timeout retry + env overrides for subscription streaming (#3248) (6ee57a509)
• feat(skills): surface reputation badges in webui (#3241) (9c3649e19)
• feat(profiles): add dynamic web interaction functions to computer-use and browser-use profiles (#3237) (d988b3053)
• feat(security): add --injection-hygiene CLI flag; screen local file reads (#3218) (bda4c1a3f)
• feat(prompts): add brief response mode (#3230) (d9be279cf)
• feat(subagent): add SubagentBudget for fleet-wide token budget coordination (#3198) (42e53289d)
• feat(tauri): add GPTME_SERVER_PORT env var to override default port (#3220) (7cd71f406)
• feat(gptme-util): add chats fork for point-in-time session forking (#3209) (8903a783d)
• feat(context): proactive context-window summarization before hard limit (#3203) (75bd94ae7)
• feat(webui): show external sessions in chat sidebar (#3219) (c20cd200f)
• feat(security): extend injection screening — shell/mcp coverage, block mode, JSONL logging (#3215) (e47142bca)
• feat(subagent): worktree isolation string API + smart cleanup (#3200) (1541032f3)
• feat(tauri): add auto-updater for desktop app (#3197) (7969b1cd4)
• feat(computer): add triple_click action and fill_native helper for native field filling (#3201) (434167ef0)
• feat(ci): weekly check for stale OpenRouter model IDs (#3204) (b16c492ad)
• feat(subagent): cancel_on_failure — cancel the fleet when a sibling fails (#3199) (214ecfdd9)
• feat(cli): compact log output for user-facing CLIs (#3195) (94122d68d)
• feat(tui): add Textual-based TUI (gptme-tui) (#3188) (c0abe9c67)
• feat(ci): add mypy type coverage reporting to Codecov (#3189) (ac1515739)
• feat(cli): add gear autonomy presets (#3180) (bfdcef7db)

🐛 Fixes (35)

Click to expand

• fix(acp): advertise authMethods in initialize response for ACP registry (#3274) (5d16f2dce)
• fix(llm): capture usage tokens from subscription response.done events (#3273) (a3ea7fdc5)
• fix(test): kill the leaked-thread teardown race class (moving "dictionary changed size during iteration" flake) (#3257) (0c8a90891)
• fix(security): replace last bare 'git' subprocess call in tasks_api.py with GIT_CMD (#3269) (3e5147f4a)
• fix(security): extend GIT_CMD hardening to all remaining bare git subprocess call sites (#3267) (36ce99eb0)
• fix(logmanager): don't merge adjacent tool-result system messages (#3268) (5be4ccf03)
• fix(security): harden bare git subprocess resolution on native Windows (#3266) (eb4be1ad1)
• fix(subagent): eliminate cancel status race — set result before writing control file (#3263) (39881fba8)
• fix(webui): latch demo mode so SPA URL rewrites can't split demo state (#3252) (16d29171c)
• fix(subagent): stop orphaned subagents at parent session end via SESSION_END hook (#3260) (3e13dd38e)
• fix(test): suppress pytest-retry StashKey teardown errors (#3256

…

These are the release notes for gptme version v0.32.0.

Contributors

Thanks to everyone who contributed to this release:

@0xbrayo, @dvirarad, @erikbjare, @Fsmount, @meichuanyi, @Miyou, @patriklaurell, @sebastiondev, @the-eater, @Thibaultjaigu, @TimeToBuildBob

Changelog

Changes since v0.31.0:

📦 gptme

✨ Features (695)

• feat(eval): add SWE-bench two-stage file retrieval (#3179) (593bdb760)
• feat(cli): add opt-in tool-output pruning (#3181) (ffee326cc)
• feat(tts): better error guidance and docs when OPENROUTER_API_KEY is absent (#3175) (3d8505cdd)
• feat(mcp): add 'gptme-util mcp serve' subcommand and docs (#3170) (80b8a76e5)
• feat(eval): wire tokens_per_task through SWE-bench and model summary (#3176) (1b71d9cd0)
• feat(models): add GPT-5.6 Luna/Terra/Sol to OpenAI model registry (#3166) (8221940df)
• feat(subagent): accept plain dict as output_schema, generate proper JSON Schema hints (#554) (#3174) (3b9fb6c6c)
• feat(eval): add tokens_consumed_per_task metric to eval harness output (#3165) (e867f7ac2)
• feat(mcp): expose gptme tools as MCP server (gptme-mcp-server) (#3157) (4de53ec1b)
• feat(acp): add gptme-acp shim package for ACP agent registry (#3160) (10adb1181)
• feat(cli): add GPTME_EXIT_STATS for machine-readable session cost summary (#3159) (2c62b349f)
• feat(computer): add --milestone and --all flags to demo command (#216) (#3152) (99621c13a)
• feat(subagent): subprocess-mode progress notifications via file channel (#3153) (3703b2813)
• feat(util/replay): compound query from last N user turns for better context recovery (#3148) (3eed3fe21)
• feat(rag): add project-scoped indexes (#3150) (08dbb2ea7)
• feat(subagent): add context_window to subagent_batch, subagent_parallel, subagent_pipeline (#554) (#3147) (3014f7b0b)
• feat(computer): add ComputerPreview panel using screenshot polling API (#216) (#3143) (ef5fe719a)
• feat(util): BM25 evidence replay for long-session context recovery (#3146) (0fa14d354)
• feat(server): add /api/v2/computer/screenshot polling API (#216) (#3140) (201fcab38)
• feat(llm): enable strict tool schema for openai and azure providers (#3136) (375008e97)
• feat(computer): add gptme-util computer demo command (#216) (#3137) (b4472da86)
• feat(subagent): add subagent_wait_any() for race/hedging patterns (7f160ba08)
• feat(computer): surface terminal startup delay fix in doctor (#216) (#3134) (31bd7b7f8)
• feat(computer): add terminal startup latency diagnostic (#216) (#3130) (338170adc)
• feat(eval): add 'Can it play Factorio?' local milestone validation (#216) (#3127) (2ea863160)
• feat(computer): add gptme-util computer doctor health-check command (#216) (#3120) (b63c9b904)
• feat(eval): add 'Can it play Doom?' local milestone validation (#216) (#3115) (006c3ef4f)
• feat(eval): add 'Can it Tweet?' local milestone validation (#216) (#3112) (e2f9f87bd)
• feat(subagent): add subagent_pipeline() for staged fan-out with no barrier (#554) (#3110) (41bb73e21)
• feat(browser): add load_browser_state() for in-session auth state restoration (#3108) (d673fb256)
• feat(browser): add hover_element(), snapshot_page(), get_current_url() (#216) (#3104) (9efc6b080)
• feat(eval): add eval tests for press_key/select_option/wait_for_element (#216) (#3097) (5f2398c9e)
• feat(subagent): token budget tracking — input/output tokens in ReturnType and BatchJob (#554) (#3099) (9de8a92cf)

…

These are the release notes for gptme version v0.31.0.

Contributors

Thanks to everyone who contributed to this release:

@0xbrayo, @erikbjare, @nikongo, @TimeToBuildBob

Changelog

Changes since v0.29.0:

📦 gptme

✨ Features (51)

• feat(lessons): auto-discover lessons from plugins (#944) (ab531ad20)
• feat(message): add MessageMetadata TypedDict for token/cost tracking (#943) (684bbb6ff)
• feat(cost): implement cost_awareness hook for session cost tracking (#939) (ed409cb26)
• feat(telemetry): improve trace quality with context propagation and rich metrics (#942) (5be5a39d4)
• feat(shell): add background job support for long-running commands (#902) (c48bf9bd7)
• feat(setup): validate API keys before saving (Issue #930) (#931) (bf1bc716b)
• feat(lessons): add caching and deduplication to lesson index (#928) (3fdbf4186)
• feat(hooks): add form auto-detection hook (Issue #591) (#919) (40dc68f45)
• feat(logmanager): add content-addressable file storage (Issue #150) (#913) (86febf7de)
• feat(tools): add form tool for structured user input (Issue #591) (#911) (92c475a47)
• feat: add Docker container for GitHub bot (Issue #305) (#905) (ec1204fa9)
• feat(tmux): add wait command to monitor long-running commands (#901) (0ee75e544)
• feat(prompt): add multi-line input support (#899) (558ee3b5f)
• feat(autocompact): add Phase 3 extractive compression for long messages (#886) (cca057c63)
• feat: add diagnostic logging to shell tool for Issue #408 (#890) (750009193)
• feat: support GPTME_TTS_SPEED and set exclude-newer for tts_server.py script to fix issues (eed3a987c)
• feat(llm): add constrained decoding support (#776) (94621288d)
• feat: add Cursor .mdc rules support (Issue #686 Phase 5) (#882) (59723c493)
• feat: Enhanced plugin management with smart src/ layout discovery (#873) (0ccdfbe44)
• feat: add compression analysis utilities and script (#864) (fd3d8edc2)
• feat(eval): GEPA Week 3 - HybridOptimizer with adaptive multi-stage optimization (#859) (58cce70f8)
• feat(context): implement hooks-based context compression architecture (#844) (20435f324)
• feat(context): implement Phase 3.1.1 core infrastructure (#860) (b898c8550)
• feat: implement active context discovery (#856) (990fdca36)
• feat: add restart tool (#853) (418532ea8)
• feat(context): Phase 1 - Core Context Selector Infrastructure (#831) (71416f6ca)
• feat: convert default model and hook registry to ContextVar for thread safety (#848) (5a1ac7b95)
• feat(plugins): implement Phase 3 command integration (#846) (c7304c43b)
• feat(plugins): implement Phase 2 hook integration (#845) (e86fe8e08)
• feat(plugins): implement Phase 1 plugin infrastructure (#843) (fe8b355cc)
• feat(server): add API v2 client for programmatic interaction (#841) (dfd9f74f1)
• feat: auto-enable complete tool in non-interactive mode and support multiple -t flags (#836) (a5ccb6cd0)
• feat(gepa): Phase 2 test set expansion - 7 more tasks (#833) (7bbce0671)
• feat(gepa): implement Phase 1 test set expansion (12 new tasks) (#832) (68bd60cfc)
• feat: support OPENROUTER_API_KEY for perplexity (#828) (658c4450a)
• feat(validation): add MESSAGE_POST_PROCESS hook for markdown codeblock cut-off detection (#824) (ed01e1cb4)
• feat: add support for gptme.local.toml configuration layering (#617) (3dfcc9e5c)
• feat(gepa): add PromptImprovementModule to fix InputField architecture issue (#823

…

v0.30.0

These are the release notes for gptme version v0.30.0.

Contributors

Thanks to everyone who contributed to this release:

@0xbrayo, @erikbjare, @TimeToBuildBob

Changelog

Changes since v0.29.0:

📦 gptme

✨ Features (26)

• feat(context): Phase 1 - Core Context Selector Infrastructure (#831) (71416f6c)
• feat: convert default model and hook registry to ContextVar for thread safety (#848) (5a1ac7b9)
• feat(plugins): implement Phase 3 command integration (#846) (c7304c43)
• feat(plugins): implement Phase 2 hook integration (#845) (e86fe8e0)
• feat(plugins): implement Phase 1 plugin infrastructure (#843) (fe8b355c)
• feat(server): add API v2 client for programmatic interaction (#841) (dfd9f74f)
• feat: auto-enable complete tool in non-interactive mode and support multiple -t flags (#836) (a5ccb6cd)
• feat(gepa): Phase 2 test set expansion - 7 more tasks (#833) (7bbce067)
• feat(gepa): implement Phase 1 test set expansion (12 new tasks) (#832) (68bd60cf)
• feat: support OPENROUTER_API_KEY for perplexity (#828) (658c4450)
• feat(validation): add MESSAGE_POST_PROCESS hook for markdown codeblock cut-off detection (#824) (ed01e1cb)
• feat: add support for gptme.local.toml configuration layering (#617) (3dfcc9e5)
• feat(gepa): add PromptImprovementModule to fix InputField architecture issue (#823) (825d71aa)
• feat(lessons): Phase 5.5 - Dynamic top-K selection (#820) (7d7fcc45)
• feat(lessons): Add ACE-inspired hybrid lesson matching (#817) (1bdd9de5)
• feat: add hook support to server API v2 (#769) (104aacff)
• feat(llm): add custom OpenAI-compatible providers support (#800) (cdf548ee)
• feat(subagent): add planner mode for task delegation (#753) (71b72b9b)
• feat(server): auto-generate auth token and document security risks (#803) (f652e752)
• feat(hooks): improve typing for hook registration with Protocol overloads (#801) (7bb17df7)
• feat(eval): implement Docker-based execution environment (#791) (7c676168)
• feat(dspy): implement multi-stage reasoning program for GEPA (#786) (a7f87c58)
• feat: log a warning when context command output is large (#787) (cf741f84)
• feat(server): implement token-based authentication for dev environment (#782) (65507d7b)
• feat(lessons): add Cursor rules integration and project-local lessons support (#779) (2ddc0e00)
• feat(tools/shell): store truncated output before discarding (#775) (b7c44982)

🐛 Fixes (30)

Click to expand

• fix: added SC2016 as excluded shellcheck code (436b335d)
• fix: fix broken favicon in the root path (#847) (421194a1)
• fix: remove xfail markers from previously flaky server tests (#849) (1645926d)
• fix(gepa): Phase 3.3 - Fix task source and auto parameter conflicts (#837) (dbbbc849)
• fix(security): block command injection via pipe-to-shell patterns (#840) (87cd354d)
• fix(gepa): remove auto parameter conflict in MIPROv2 (#835) (7e51e0ac)
• fix: fixed Kimi K2 thinking toolcall support via OpenRouter (#830) (5a758dbc)
• fix: resolve MCP tool loading issues and connection errors (#825) (72a4d85b)
• fix(subagent): add missing tool_format parameter to chat() calls (3167b336)
• fix(gepa): export ANTHROPIC_API_KEY and suppress verbose logs (#821) (aeefe9b4)
• fix(gepa): collect and save trajectory feedback in optimization results (#819) (603ccfef)
• fix(eval): fix reflection model used in GEPA optimizer (#814) (6345f0c2)
• fix(eval): add defensive check for eval_result in task_success_metric (#813) (644eadd9)
• fix(server): add GPTME_DISABLE_AUTH env var for k8s deployments (#811) (dd491017)

…

v0.29.0

These are the release notes for gptme version v0.29.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare, @TimeToBuildBob

Changelog

Changes since v0.28.3:

📦 gptme

✨ Features (39)

• feat: improve token awareness to show warnings at meaningful thresholds (fce2382a)
• feat: make hook parameters consistent across all hook types (2e71dcc9)
• feat: expose MCP tool commands as slash commands (#760) (2a89022f)
• feat(mcp): add MCP discovery and dynamic loading support (#685) (e8af7604)
• feat: add token & time awareness tools/hooks (#665) (c0391ddd)
• feat: show conversation ID in goodbye message (ad5bc4d0)
• feat: add gptme-util prompts expand command (fb543142)
• feat: add --profile flag for performance profiling (9ce6e3e8)
• feat(shell): add configurable shellcheck exclude codes (#747) (46a54148)
• feat(shell): add simple shellcheck validation (#740) (20f28f6b)
• feat(lessons): add message pattern-based mode detection for Phase 3 (#722) (fc131775)
• feat: switch CI from claude-3-5-haiku to claude-haiku-4-5 (#712) (64e04708)
• feat(llm): add support for Claude Haiku 4.5 (#705) (e48fcd2f)
• feat(lessons): implement status filtering for lesson lifecycle management (#702) (be25b397)
• feat: support @log workspace magic value in API (#695) (9fc71ad1)
• feat: save removed tool outputs during auto-compacting (186ac031)
• feat(gh): add optional commit_sha parameter to check specific commits (3df3c44f)
• feat(lessons): add configurable lesson directories via gptme.toml (0a05ec61)
• feat(gh): add quick status check command (76c67881)
• feat(gh): show run IDs for failed checks and add --log-failed example (c3f131c3)
• feat: enhance lesson matching with tools and history-based deduplication (5aefbe37)
• feat(llm): add configurable API timeout for OpenAI-compatible providers (#681) (6a2d532f)
• feat: enhance GitHub PR reading with code context and suggestions (#689) (4c1cef43)
• feat: implement basic lesson system with metadata and auto-inclusion (#687) (3e81a0d2)
• feat: add tool_format to telemetry and remove redundant log (d59332ab)
• feat(browser): add Accept header to prefer markdown/plaintext over HTML (#680) (f92ab671)
• feat: improve conversation history context (#679) (35c303d3)
• feat(shell): add 20-minute default timeout to prevent stuck commands (#678) (526da1cb)
• feat(telemetry): add hostname to resource attributes (#676) (0d77253f)
• feat(telemetry): add agent name and interactive mode metadata (#675) (206ff1fb)
• feat: reintroduce OpenAI and Anthropic telemetry instrumentation (d4b031f0)
• feat: add GENERATION_PRE hook trigger before LLM generation (f69c3d37)
• feat: add concise repr to Log class (b9485b98)
• feat: implement auto-reply mechanism as LOOP_CONTINUE hook (de79d5ba)
• feat: implement /compact and auto-compacting (#645) (61392fc0)
• feat: implement comprehensive hook system and command registration (#156) (#660) (32d39817)
• feat: auto-restore todo state when resuming conversations (#659) (debcb096)
• feat: implement complete tool and auto-reply mechanism for autonomous operation (14952de1)
• feat: emit hint when save tool barely changes file (1601f5e8)

🐛 Fixes (77)

Click to expand

• fix: fix issue with hook args (4fa95612)
• fix: correct complete_hook signature to match GENERATION_PRE calling convention (90a12f9c)
• fix: handle pipe operators inside quoted strings in shell commands (#764) (5f80b533)
• fix: skip generating initial messages when resuming conversations (#763) (2898e79d)
• fix: add special haiku-adapted to_xml (for testing) (cc868960)
• fix(tools): add support for Haiku 4.5 XML format (#752) (96f87e43)
• fix: run auto-naming in background thread to avoid blocking CLI (1d17674f)

…

These are the release notes for gptme version v0.28.3.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare, @TimeToBuildBob

Changelog

Changes since v0.28.2:

📦 gptme

✨ Features (2)

• feat: improve save tool feedback with detailed status (d920ae5e)
• feat: add git safety guards to shell tool (fc78a042)

🐛 Fixes (7)

Click to expand

• fix: update CI anthropic model to claude-3-5-haiku (28d9367c)
• fix(shell): denylist should not trigger on content in quoted strings or heredocs (0ce788a2)
• fix(shell): improve denylist patterns for git commands (5e6bdc97)
• fix: prevent premature code block closure during streaming with nested blocks (#657) (55fedb76)
• fix: change default/recommended model to Sonnet 4.5 (21de08ce)
• fix: use dateutil.parser.isoparse for all datetime parsing (f6f3de5e)
• fix: use dateutil.parser.isoparse instead of datetime.fromisoformat (934384dc)

🔨 Misc (5)

• chore: bump version to 0.28.3 (d22ff504)
• docs: fixed v0.1.1 release notes (00ed0ece)
• docs: included all past changelogs (6d407047)
• docs: improve chat history prompt format (6fbd9db8)
• docs: added release notes for v0.28.2 (481a07ff)

(excluded 2 less relevant commits)

Full Changelog: https://github.com/gptme/gptme/compare/v0.28.2...v0.28.3

v0.28.2

These are the release notes for gptme version v0.28.2.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.28.1:

📦 gptme

✨ Features (3)

• feat: add dry-run mode for DSPy optimization commands (bd8861c5)
• feat: add dedicated gptme-dspy command entry point (4c400d9f)
• feat: GEPA in DSPy (#651) (014121ce)

🐛 Fixes (5)

• fix: add support for sonnet 4.5 (991e8ac0)
• fix: fixes to proxy openrouter support (7f4e7bf7)
• fix(llm): respect LLM_PROXY_URL for OpenRouter models endpoint (17b903c9)
• fix: make num_trials parameter actually control DSPy optimizers and add CLI options for dataset sizes (c447ed5c)
• fix: improvements to evals/dspy/gepa (#652) (a1a17ea5)

🔨 Misc (5)

• chore: bump version to 0.28.2 (3b7d5fc5)
• refactor: split GEPA into separate optimize-gepa subcommand with proper budget configuration (67588b16)
• docs: added release notes to index (4229c49a)
• docs: added release notes for v0.28.1 (9836de12)
• chore: updated changelog_contributors.csv cache (41f40093)

Full Changelog: https://github.com/gptme/gptme/compare/v0.28.1...v0.28.2

These are the release notes for gptme version v0.28.1.

Contributors

Thanks to everyone who contributed to this release:

@delorenj, @erikbjare

Changelog

Changes since v0.28.0:

📦 gptme

✨ Features (14)

• feat: add denylist for dangerous shell commands with specific deny reasons (#648) (5b977846)
• feat: implement get_model TODO and fix critical telemetry bug (#647) (9c05ac26)
• feat: add trajectory-rich tasks for GEPA optimization (#640) (962bdf3c)
• feat: add user-level files support in config (#639) (fb3bbfb7)
• feat: add HTTP MCP Server Support + CLI Management Tools (#635) (0a4ef349)
• feat: added cross-conversation context (#636) (54f289ec)
• feat: add cache-aware cost calculation, use in telemetry (#631) (b3cd6e07)
• feat: dspy experiment (#627) (fb61ddda)
• feat: filter out resolved comments from GitHub PR content (d612a355)
• feat: add GitHub Actions status to PR content fetching (947700de)
• feat: add auto-naming for conversations in server API (#621) (cfce2bc8)
• feat: add todo tools for conversation-scoped task management (#622) (ce3bfe72)
• feat: include pwd and agent workspace path in system prompt (b38e06d0)
• feat: add GitHub issue/PR link handling in context (#619) (f79e2428)

🐛 Fixes (25)

Click to expand

• fix(api): support auto-generating agent path from name (#646) (62534a95)
• fix: propagate agent logdir from subprocess in DSPy evaluations (#643) (d6c9fc69)
• fix(eval): fix DSPy integration to use real evaluation specs (#630) (fa48bb34)
• fix: added qwen3-max metadata (a471d1de)
• fix: add initial support for magistral models (59dce119)
• fix: fixes on top of HTTP MCP server support (#637) (21dc660a)
• fix: switch to sonar-pro by default for perplexity (d9e99aef)
• fix: extract env vars to constants (0eb0c9b0)
• fix: improve todo tool, support writing multiple tasks in one tool call (61727b43)
• fix: correct optimizers parameter format in workflow (9c565269)
• fix: handle models that dont support vision, stricter openrouter provider selection, fixes to deepseek & kimi-k2 (1c18c3fa)
• fix: added chime to generate_sounds script (fe24000f)
• fix: misc fixes (#626) (ff0b6c74)
• fix: catch exception if image file cannot be read (8165857c)
• fix: update provider examples (faf56f60)
• fix: move GitHub Actions status to end of PR content (174cc409)
• fix: make morph tool work with all tool formats (fixes #603) (181a5fca)
• fix: better tree output (#624) (6782d81a)
• fix: fixed todo tools (daf128aa)
• fix: add browser resilience with auto-restart and retry mechanism (26c93942)
• fix: include starting working directory and use absolute paths in system prompt (204d09d2)
• fix(server): fix setting initial working directory from workspace (03aa619e)
• fix: use "Initial Working Directory" instead of "Current Directory" (c6cbc272)
• fix: exclude untracked files from autocommit status check (da5fb7f5)
• fix: resolve audio blocking and ALSA/PulseAudio timeout issues (#620) (d6703e85)

🔨 Misc (13)

Click to expand

• chore: bump version to 0.28.1 (dae71fc5)
• test: fix todo tool test expectations to match implementation (e9ff2efc)
• docs: fixed incorrect example (1aaad3d0)
• chore: comment out excessive debug logging (65f68e23)
• refactor: improve ask_execute variable naming and imports (471ad1b5)
• docs: clarify gptme description and improve usage examples (f606ccd2)
• docs: fix outdated model reference (d7a06264)
• chore: fix gitignore (8e262818)
• docs: fix docs for changelog/release notes (#623) (8c6caf5f)
• docs: create changelogs for tagged versions in ./docs/changelog using make target (4f69ef2e)
• docs: update project description to emphasize modern agent capabilities (37a179a1)

…

These are the release notes for gptme v0.28.0.

Usage

To get started, see the Getting Started guide in the documentation.

Contributors

Thanks to everyone who contributed to this release:

@0xbrayo, @erikbjare, @Miyou, @RickardCarlsson4

Changelog

Changes since v0.27.0:

Summary

This is a major feature release packed with exciting improvements! 🎉

🚀 Major New Capabilities:

• GPT-5 Support - Works with OpenAI's latest model
• Claude 4 Support - Works with Anthropic's latest model
• MCP Integration 🔌 - Support for Model Context Protocol servers, expanding tool ecosystem

🤖 Agent & Automation Enhancements:

• Auto-commit - Automatically tell agent to git commit, or do it manually with /commit
• Custom Agent Improvements - Even better support for agents based on the gptme-agent-template
• Agent Workspace Auto-detection - Smarter handling of agent environments

🛠 Tool Improvements:

• Shell Timeouts ⏱️ - Configurable command timeouts (set GPTME_SHELL_TIMEOUT)
• Morph Tool ✨ - Alternative fast patching tool for precise code edits
• Perplexity Search 🔍 - Enhanced web search capabilities through the browser tool
• Computer Tool Improvements 💻 - Better key sequence chaining for GUI automation
• Enhanced TTS 🗣️ - Improved text-to-speech with Chatterbox support

🌐 Server & API:

• Redesigned Server API - Complete API overhaul for better web UI integration
• Tasks API - New task management capabilities
• OpenAPI Specification - Comprehensive API documentation

🔧 Quality of Life:

• Setup Command ⚙️ - Easy configuration with /setup
• 90+ Bug Fixes - Significantly improved stability and reliability
• Better Configuration - Enhanced config system with user-, project-, and chat-specific settings
• Improved Error Handling - More user-friendly error messages and recovery
• Bell and Tool Sounds 🔊 - Pleasant notification sounds for user input requested and different operations (enable via /setup or GPTME_DING/GPTME_TOOL_SOUNDS env vars)

📦 gptme

✨ Features (39)

• feat: enhance telemetry with Prometheus metrics and comprehensive instrumentation (#592) (a73f60b0)
• feat: improve eval system and fix tool format content extraction (#618) (099c80b4)
• feat: add GPT-5 support (4ff6d5c9)
• feat: add timeout support to shell tool (06d85f94)
• feat: add last/all option to /replay command (ec4bec77)
• feat: added /api/v2/models endpoint (#612) (b08f64e0)
• feat: add tool sounds for different operations (#597) (ce6f50a2)
• feat: add Perplexity search support to browser tool (#566) (c91c9d18)
• feat: implemented choice tool (#598) (07b602ee)
• feat: add /setup command for user and project configuration (#596) (2379e665)
• feat: add ding sound notification with bell audio (#588) (6096474b)
• feat: add OpenTelemetry integration for performance monitoring (#587) (41ab4283)
• feat: add support for custom agent names in CLI/TUI interface (658fd3d6)
• feat: auto-detect agent workspaces (420d3072)
• feat: added support for --agent-path in CLI, include agent context in addition to workspace context (c40477e0)
• feat: implement auto-commit (#441) (cf0ed1c5)
• feat: support the openrouter /v1/models api to list all models with metadata (#575) (88df6238)
• feat: added alternate patch tool using morph fast apply v2 via openro… (#574) (23258e0b)
• feat: add support for a GPTME_REASONING flag to explicitly enable/disable reasoning (#568) (df0e0ca7)
• feat: add automatic inclusion of default project files (#565) (ac05da57)
• feat: add chatterbox tts support (#541) (84151f97)
• feat: openapi spec for server (#563) (62bd55d6)
• feat: add tasks API for new tasks UI in gptme-webui (#562) (6807d484)
• feat: added 'worked for <time>' output behind feature flag (due to issues) (ce78dda1)
• feat: added read_logs function to browser tool (playwright backend) (#553) (9291e000)
• feat: add gptme-util models list/info commands with fish completion (82c551f7)
• feat: add support for setting chat name in chat config (#549) (5c30f5df)
• feat: added gptme-util llm generate command, rework gptme-util context (#535) (87832050)
• feat: added basic workspace file browsing api (#530) (0ab8abf9)

…

These are the release notes for gptme version v0.27.0.

Contributors

Thanks to everyone who contributed to this release:

@0xbrayo, @erikbjare, @jrmi, @Miyou

✨ Summary

This release brings a wealth of exciting improvements and new capabilities to gptme:

🚀 Major Enhancements

• 🔧 Pre-commit Integration: Automatically run pre-commit hooks when tool use is exhausted, keeping your code clean and consistent!
• 🍎 macOS Computer Use Tool: Extended our powerful computer use tool to macOS users, allowing the assistant to interact with GUI applications
• 🧠 Reasoning Models Better Supported:
  • Claude 3.7 Sonnet support
  • DeepSeek R1 support
• 🔈 Improved TTS: Enhanced Text-to-Speech with Kokoro 1.0, better sentence chunking and device handling

🛠️ Developer Experience

• 🌳 Context Tree: Added GPTME_CONTEXT_TREE flag to include repository structure in prompts
• ⚙️ Feature Flags: New environment variables like GPTME_BREAK_ON_TOOLUSE and GPTME_FRESH for advanced users
• 🔍 Enhanced RAG: Much improved document retrieval with LLM post-processing
• 📊 Terminal Status: Real-time status updates in terminal title
• 🔄 Auto-rename: Added script to intelligently rename past conversations

🔧 Under the Hood

• 🐛 Bug Fixes: Fixed 65+ issues to ensure smooth, reliable operation
• 📷 Better Image Handling: Improved vision tool with automatic rescaling
• 📋 Model Switching: Added /model command to list and switch models in conversation
• 🔄 Patch Recovery: New GPTME_PATCH_RECOVERY option to recover from non-matching patches

We apologize for the delay in this release. For those eager to try the latest features before the next release, we recommend running directly from the master branch:

pipx install git+https://github.com/gptme/gptme.git

Changelog

Changes since v0.26.0:

📦 gptme

✨ Features (19)

• feat: added GPTME_CONTEXT_TREE which includes tree --gitignore . output in initial prompt (#461) (593ee900)
• feat(style): print <think>/<thinking> in dimmed text (dfab1e05)
• feat: added basic files support to server API (#440) (24b9dc95)
• feat: added script to auto-rename past conversations (#439) (9a9f0114)
• feat: added terminal title status output (#438) (37c877f3)
• feat: much improved RAG, added LLM post-processing of results (#435) (f4b6f02e)
• feat: use gnome-screenshot in wayland (#436) (6d1d47a4)
• feat: add status emojis to system messages (#429) (69c5f011)
• feat: added support for nvidia provider (7f387d33)
• feat: auto-enable pre-commit checks when config exists (#430) (5cd9b1e8)
• feat: enhance vision tool with better image processing (6a982987)
• feat: add option to override base prompt (#426) (78e92091)
• feat(computer): add macOS support for computer tool (#324) (7a08e341)
• feat: added support for GPTME_BREAK_ON_TOOLUSE flag to not stop generation when tooluse occurs in stream (e7021b9c)
• feat: auto-run precommit checks when tooluse exhausted (#415) (254d1b61)
• feat: added support for deepseek-reasoner (#410) (ac3b8258)
• feat: added /model command to list/switch models, support mixing providers (#409) (a2bc4e00)
• feat: add env var feature flag for llm prompt suggestions (978b1c4b)
• feat: added gptme-wut (#402) (91bdce67)

🐛 Fixes (65)

Click to expand

• fix: fixed reduce_context script (2f8246c6)
• fix: added reduce_context.py script (f0edadde)
• fix: added gh-pr-view-with-pr-comments.sh script (3e0dd23d)
• fix: add gemini-2.0-flash-thinking-exp-01-21 to models (#465) (489f73e6)
• fix: fix /export command (#460) (0ce55149)
• fix: fix missing dependency warning (#455) (ced1b9ac)
• fix: fixed incorrect reference to GPTME_FRESH_CONTEXT (now just GPTME_FRESH) (ce626ed1)
• fix: fix circular import (417613c1)
• fix: correctly handle too long path names (ec1118c6)
• fix: dont expand paths in user commands for tools, like /shell (#453) (96304695)
• fix: fixed a bunch of get_default_model uses, return None instead of exception if unset (a3626c82)
• fix: make model fully optional in prompts.py, no fallback to possibly unset default (29a0cb19)
• fix: correct content block processing in Anthropic LLM module (b4d2d962)
• fix(anthropic): use model_meta.supports_reasoning to determine wether to use thinking (91af56c1)
• fix: add supports_reasoning to model_meta, disabling <thinking> prompting for such models (b15fb6a5)

…

These are the release notes for gptme version v0.26.0.

Summary

This release includes:

• Support for local TTS using Kokoro (experimental, #401, announcement tweet)
• Ability to load tools from external modules (#336)
• Many fixes and improvements to reliability.

Contributors

Thanks to everyone who contributed to this release:

@0xbrayo, @chenrui333, @erikbjare, @jrmi

Changelog

Changes since v0.25.0:

📦 gptme

✨ Features (4)

• feat: added local TTS support using Kokoro (#401) (3e8e8698)
• feat: allow to load tools from external modules (#336) (dd99731c)
• feat: add support for different tool formats in eval system (#380) (5fbc0acb)
• feat: improve tool use response format (#306) (1934fcf1)

🐛 Fixes (16)

Click to expand

• fix: fixed sending whitespace-only messages to tts server (51ce6aea)
• fix: more precise version when installed in editable mode, with git info (#400) (9cbaa5da)
• fix: allow overriding logs folder via GPTME_LOGS_HOME (a34276c7)
• fix: dont store invalid commands in log (aaf2e9e9)
• fix: fix /summarize command (#390) (c41df98e)
• fix(server): init tools in server api (#389) (b6f337ac)
• fix: tool api call broken when user answer no to when asked for confirmation (#371) (72fb003d)
• fix: several fixes to tmux tool, incl support for multiple commands per tooluse (#382) (8dbed8eb)
• fix: rename python tool to ipython for better tooluse format adherence (#361) (d4407e30)
• fix: improve smaller model response on user refusals (#374) (11708dd1)
• fix: create workspace symlink in log folder, load on resume (#368) (03475102)
• fix: shorten command descriptions (0d642c22)
• fix: remove conflicting -n alias for --name, keep as --non-interactive alias (fixes #360) (dd26dc23)
• fix: prettier /command completions (f2e01f3b)
• fix: retry on anthropic overloaded (#356) (50763762)
• fix: fixed auto n option on ask_execute, added log message with remaining skips (cc1af13a)

🔨 Misc (10)

Click to expand

• chore: bump version to 0.26.0 (3fef9941)
• chore: fixed venv docker/gitignore (a32fc3bc)
• docs: improve custom tools docs, mentioning simple script tools (#391) (d33ff353)
• test: simplified tool format test, dont skip cli url test if lynx available (cdbfae8b)
• chore: added .aider and .env to gitignore (54253a5d)
• test: fix reset_default_model fixture if no default set (a5687952)
• test: reset default model in openai tests to fix flaky test (7ff84922)
• test: fix typo in model name (dec7a60f)
• chore: update brayo-pip -> 0xbrayo (871cf63c)
• chore: updated changelog_contributors.csv cache (7abeb1eb)

(excluded 11 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.25.0...v0.26.0

These are the release notes for gptme version v0.25.0.

Summary

This release brings major improvements to the user experience with a new prompt_toolkit-based interface (better completion and highlighting), support for OpenAI/Anthropic tools APIs, and significant cost & performance improvements for Anthropic through better prompt caching. The release also includes improved path handling, workspace context, heredoc support, and numerous UX improvements and bug fixes.

Contributors

Thanks to everyone who contributed to this release:

@jrmi, @bjsi, @0xbrayo, @erikbjare

Changelog

Changes since v0.24.1:

📦 gptme

✨ Features (8)

• feat: support auto n to skip n confirmations (#346) (7d2e9e7b)
• feat: migrate readline to prompt_toolkit, with many features and fixes (#244) (61ca0a24)
• feat(llm): add model metadata for Deepseek, Groq, and XAI (d2109fef)
• feat: simplify rag tool by simply calling gptme-rag via subprocess (#316) (87571086)
• feat(anthropic): improve prompt caching and type safety (#317) (a2f06df8)
• feat: started working on always-fresh context message (#281) (01d80524)
• feat: support for OpenAI/Anthropic tools APIs (#300) (d83f7238)
• feat: output costs for request and session (#293) (063624fe)

🐛 Fixes (34)

Click to expand

• fix: return instructions to correctly playwright install if browser binaries unavailable (#358) (ebf56754)
• fix: dont log costs by default (f65e75e6)
• fix: stream and capture ipython output (#357) (ef424bed)
• fix: removed spammy debug-logging message (99912f3c)
• fix: properly handle piped input in prompts (#354) (f8321834)
• fix: support piping stdin into gptme -r (resume) (#344) (621ffd80)
• fix: move playwright to thread to avoid event loop conflicts (#353) (803d6e51)
• fix: migrated to latest anthropic version v0.42 (prompt caching now stable) (#352) (0ecf0459)
• fix: fix reading stdin after prompt-toolkit migration (#337) (b3974c15)
• fix: fix log output during input, fix directory completion (#338) (e3874aa4)
• fix: remove prompting to avoid the now-supported EOF/HereDoc (3950a6e7)
• fix: support heredoc/EOF syntax in shell tool (#335) (6a1246f5)
• fix: improved path detection in prompts, now works with more adjacent punctuation types (#333) (8e8f28ea)
• fix: use full paths to files in workspace context prompt (#330) (a61aa284)
• fix: fixed incorrectly always using gpt-4 model metadata when reducing (2343ab67)
• fix: when patch failed due to bad path, include pwd in message for fast recovery (202869e6)
• fix(llm): transform message content for Groq provider (84c83166)
• fix: better prompt caching & less debug logging (#323) (fae65ca4)
• fix: better anthropic cost logging (#321) (283e8b0f)
• fix: better error output when loading locked conversation (#319) (2264cf10)
• fix: further anthropic caching improvements (#318) (65efc3e0)
• fix: fixed regression in syntax highlighting for saves after refactor (a40f5387)
• fix: added conversation lock file management in LogManager (#295) (7255f255)
• fix: improvements after execute_with_confirmation refactor (#311) (23f81cf8)
• fix: set TOKENIZERS_PARALLELISM env var to false to get rid of warning (75b298be)
• fix: clarify return prompt for subagent tool (29bcc473)
• fix: skip loading subagent tool by default (37ae9c19)
• fix(prompts): minor fixes and improvements to prompts (10108a10)
• fix: fixed Dockerfile for computer use (9bf733a5)
• fix: include mention of user-edited save/patch/command in system response (93e126ac)
• fix: make the --help text for --model always use recommended models (2c872f36)
• fix: handle exceptions if readline history fails to write (c60eb883)
• fix: prevent command execution from triggering unnecessary response generation (7f66dbff)
• fix: incorrect response_preference key in config (748be0b6)

🔨 Misc (26)

Click to expand

…

These are the release notes for gptme version v0.24.1.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.24.0:

📦 gptme

✨ Features (1)

• feat(llm): support Gemini provider (#272) (e875c55)

🐛 Fixes (4)

• fix(vim): improved window configuration (e59cc06)
• fix: deepseek provider and gptme-eval docker improvements (#270) (591b5be)
• fix: model provider handling and vision support metadata (#269) (7a797cf)
• fix: remove o1 streaming restriction (7fd6ad5)

🔨 Misc (15)

Click to expand

• chore: bump version to 0.24.1 (5ec053f)
• docs: added gptme.vim to projects (e3afcfd)
• refactor: moved vim plugin into its own repo ErikBjare/gptme.vim (190fcd7)
• docs(server): mention how to install server extras (cb561d6)
• docs: move detailed usage to separate page (e8ec1c8)
• docs: improve examples documentation (13e3b95)
• docs: move automation document under examples (c2b904f)
• docs: add projects page, more examples and 'powered by gptme' badge (602155e)
• docs(examples): added more examples (e69551f)
• docs: added links to changelog (releases) and X, improved examples (use of git status -vv) (37ab6b4)
• docs(README): changed twitter/x badge to point to @gptmeorg (906bc96)
• chore: added htmlcov to gitignore (cfd7cdf)
• test: fixed readline completions test (bc79656)
• docs(README): moved agent feature to done features, link to gptme-agent-template (e17ef36)
• refactor(rag): improve RAG manager implementation (#268) (63e8d04)

(excluded 7 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.24.0...v0.24.1

These are the release notes for gptme version v0.24.0.

Summary

🌟 New Features

• Added RAG (Retrieval-Augmented Generation) support for context-aware assistance (#258)
• Added gptme-util CLI with tools and chats management commands (#261, #265)
• Added current date & time to system prompt for better temporal awareness

🔧 Improvements & Fixes

• Fixed patch tool to support line removal (#267)
• Fixed path handling in patch and save tools
• Fixed save diff preview for paths starting with ~/
• Set PAGER and GH_PAGER in shell tool
• Added wait-for-ci example to gh tool

🏗️ Refactoring & Developer Experience

• Grouped all provider/model logic into llm directory (#254)
• Simplified RAG configuration and initialization (#266)
• Improved test coverage for patch tool
• Added verbose logging and tools call command
• Improved Docker build configuration
• Added pylint duplicate-code checks

Contributors

Thanks to everyone who contributed to this release:

@erikbjare, @jrmi

Changelog

Changes since v0.23.0:

📦 gptme

✨ Features (5)

• feat(cli): print return values from tool function calls (c1f4c00)
• feat(utils): add verbose logging and tools call command to gptme-util (#265) (b3955d5)
• feat: begin work on RAG, refactoring, improved docs and tests (#258) (950c397)
• feat: implement gptme-util CLI for utilities (#261) (128676c)
• feat: added current date & time to system prompt (faaec3a)

🐛 Fixes (5)

• fix: support line removal in patch tool (#267) (7c51573)
• fix: enforce full path usage in patch tool (2ade667)
• fix(save): fix save diff preview when path starts with ~/ (0fbafd0)
• fix: set PAGER and GH_PAGER in shell tool, add wait-for-ci example to gh tool (e4ce6c8)
• fix: mention paths being relative to working dir in prompt for save tool (41594c5)

🔨 Misc (8)

Click to expand

• chore: bump version to 0.24.0 (990965f)
• test(patch): improve test coverage for patch tool (e1e2943)
• test(patch): improve empty patch test (a660512)
• refactor(rag): simplify configuration and initialization (#266) (84e5ab6)
• refactor: group all provider/model related logic into llm directory (#254) (ac3c173)
• docs(README): added links to issues for agents and Bob (ebd700b)
• chore: moved mypy options into pyproject.toml (ba2009a)
• docs: moved a couple 'in progress' features to the completed features section (e9a0022)

(excluded 2 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.23.0...v0.24.0

These are the release notes for gptme version v0.23.0.

Summary

• UX improvements
  • New c option in confirmation prompt to copy to clipboard
  • New auto option in confirmation prompt (effectively the same as --no-confirm).
• Server API now supports streaming & interruptions in the new fancy web UI at https://gptme.gptengineer.run/
• Computer use improvements
• gptme.vim plugin

Contributors

Thanks to everyone who contributed to this release:

@brayo-pip, @erikbjare, @jrmi

Changelog

Changes since v0.22.0:

📦 gptme

✨ Features (6)

• feat: improved ask_execute options interface, now supporting '?' and 'auto' (66c977c)
• feat: copy to clipboard (#234) (7ad4c94)
• feat(server): support interrupting stream, support slash commands (#248) (c1d136f)
• feat: added support for streaming in API (#247) (7b62201)
• feat: added --cors-origin cli param to gptme-server (a7190ed)
• feat: implemented gptme.vim plugin (#241) (a04ee05)

🐛 Fixes (11)

Click to expand

• fix: prompt to reinforce correct placement of filename parameter in patch tool (#252) (01541af)
• fix: trigger tool detection only on complete lines, for performance and to fix nested codeblocks in tooluse (#251) (5a228d4)
• fix: fixed prompt for python tool (64f9491)
• fix: fixes to computer use (1de06c6)
• fix: switch to new Claude 3.5 models in evals, improve detection of results folders (6aec231)
• fix: switch to duckduckgo as default search engine with lynx browser (#237) (ce7afb8)
• fix: fixed running evals in docker (5ac3914)
• fix: added metadata for new haiku model (8564756)
• fix: fix including files specified in gptme.toml (381dbc8)
• fix: updated READMEs about gptme.vim (f7913bb)
• fix: removed debug logging in gptme.vim (25c998c)

🔨 Misc (12)

Click to expand

• chore: bump version to 0.23.0 (9010fed)
• chore: committed caches for contributor info used in changelog generation (9dee554)
• docs: improved server/computer use docs (501008a)
• refactor: move get_workspace_prompt into prompts.py, add TODO comments (639e4cb)
• docs: updated gptme --help output, fix command ordering (da90e33)
• docs: minor improvements to server docs (f6589de)
• docs: updated server docs with better structure and mention of gptme-webui (c32f64d)
• docs(README): moved the use cases section above the developer/in-progress sections (0c0163d)
• docs(README): rewrote use cases section (a517d8f)
• docs: add link to SWE-Bench PR (970fc47)
• docs: improved arewetiny intro text (3f5aa84)
• docs: fixed gptme.vim docs (9ca2e08)

(excluded 3 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.22.0...v0.23.0

These are the release notes for gptme version v0.22.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.21.0:

📦 gptme

✨ Features (5)

• feat: implement anthropic-style computer tool (#225) (175167e)
• feat: added /export comment to export self-contained HTML file built from webui (#235) (0ca914f)
• feat: add platform info to the system prompt (#171) (0288d33)
• feat: added lynx browser support (#214) (2cd9ffb)
• feat: added deepseek support (#180) (949eaee)

🐛 Fixes (11)

Click to expand

• fix: correct Linux distro version in systeminfo prompt (#239) (3221145)
• fix: informative message for agent when permission isn't granted (#223) (5ee0f2f)
• fix: fixed version looking up old package name (f8c296f)
• fix: set <title> in export and make sure /export command doesn't get saved to log (2a5576a)
• fix: improved chat picker (now uses full-width of terminal) (32db19d)
• fix: add cli-provided prompts to readline history (ffaa109)
• fix: improved reliability of llm.generate_name() (b931103)
• fix: include conversation name in exported chat log (abc4928)
• fix: changed all Message-generator tools to simply print/return value instead (fixes #186, fixes #187) (3dcef9f)
• fix: fixed support for groq (and deepseek?) (#231) (cea30cf)
• fix: fixed screenshot() not running if generator not consumed (such as when not last statement in codeblock) (65bdb8a)

🔨 Misc (8)

Click to expand

• chore: bump version to 0.22.0 (cdd8c6b)
• docs: removed bad computer use example (66898b2)
• tests: fixed blinking test due to changed working dir (40d1a5d)
• docs: add installing from source instructions to contributing guide (#236) (4665f49)
• format: s/whitelist/allowlist (e77cd4e)
• docs: add mention of configuration file on providers page (16f8254)
• docs: fixed incorrect local/ollama/... provider prefix (3eb6f8f)
• docs: improved note about outdated demos (cdce637)

(excluded 2 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.21.0...v0.22.0

These are the release notes for gptme version v0.21.0.

Contributors

Thanks to everyone who contributed to this release:

@brayo-pip, @erikbjare

Changelog

Changes since v0.20.0:

📦 gptme

✨ Features (4)

• feat: added support for groq provider (4299cd0)
• feat:whitelist some commands (7f1ba2b)
• feat: added support for xAI/grok (d7cebba)
• feat: started working on ncurses ui (d3413ea)

🐛 Fixes (19)

Click to expand

• fix: added 'head' to allowlisted commands in shell tool (7cc752f)
• fix: compile cmd_regex (#222) (623a52d)
• fix: print used model on startup (6cf4001)
• fix: better error if attempting to run on Windows, refactor readline stuff (#221) (bd8b746)
• fix: update to use latest Sonnet model by default, improve typing (6e70168)
• fix: dont catch interrupts until conversation has begun (2be45a8)
• fix: added ncgptme and gptme-nc script entrypoints for ncurses tui (7b19b76)
• fix: fixed typing in ncurses.py (f018c6c)
• fix: improved shell tool preview format (756e420)
• fix: change OPENAI_API_BASE to OPENAI_BASE_URL (30e3f01)
• fix: ensure subagent logdir unique (f735111)
• fix: changed subagent tool param order and example (2bdb48f)
• fix: fixed display bug in confirm prompt (b9f8cd8)
• fix: fixed bug where user prompt not included in request (53d160e)
• fix: fixed prompt_user returning empty string when interrupted (440aedb)
• fix: fixed leftover call to ask_execute instead of confirm func (b7d2a3f)
• fix: fixed incorrectly asking for confirmation when impersonating (9498343)
• fix: more fixes/improvements to treeofthoughts.py (d54df51)
• fix: re-raise tool use errors in tests (0ada191)

🔨 Misc (14)

Click to expand

• chore: bump version to 0.21.0 (852e6f4)
• docs: added more WIP features to README (2d8b602)
• docs: added note about Windows support (b8ce040)
• docs: added docstring to ncurses.py (5bb0024)
• docs: add note about limitations of small local models, convert page to rst (f521c71)
• format: fixed lints (ff2277c)
• refactor: refactor ncurses.py and add --no-color cli argument (942996c)
• docs: added basic docs for configuration files (fixes #173) (acd0cee)
• docs(README): removed old ToC link (5801bbf)
• docs: fixed docs building after refactor (52eaed9)
• refactor: refactor how confirmation works, enabling LLM-guided confirmation and simplifying confirmation support in server (b843e88)
• refactor: work on programmatic interface, refactored LogManager into mutable manager and immutable Log dataclass, added wip treeofthought script (d421cc8)
• format: fixed formatting after pre-commit setup (4ee9761)
• docs: fixed incorrect OPENAI_API_BASE url (732c5b8)

(excluded 4 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.20.0...v0.21.0

These are the release notes for gptme version v0.20.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.19.2:

📦 gptme

✨ Features (1)

• feat: updated webui style, with sidebar (bd44e5c)

🐛 Fixes (10)

Click to expand

• fix: use importlib.util.find_spec instead of attempting costly imports at startup, reducing startup time by ~1s (7b75520)
• fix(webui): put <thinking> tags into <details>, minor style improvements (04b3109)
• fix(webui): switch to marked over showdown, improved styling and misc fixes (0a868ba)
• fix: catch exceptions when executing tools (9560660)
• fix: handle bad patches better (771734c)
• fix: limit image size, fixes exception for large images (fixes #185) (#188) (45cfbac)
• fix: fixed incorrectly nested string in f-string (4aa3f2d)
• fix: made shell tool more strict, now requires exact 'shell' langtag to run (578adcc)
• fix: use stdout/stderr as langtags in shell output, instead of as headings (c102806)
• fix: log warning if allowlisted tool could not be found (f1864c7)

🔨 Misc (9)

Click to expand

• chore: bump version to 0.20.0 (5d0b138)
• docs: added startup time to 'Are we tiny?' page (cde7937)
• refactor: extracted js part of webui into seperate file (919fe93)
• docs: improved styling of demos (018e6ff)
• docs: more use of rubric::, use sphinxcontrib.asciinema to embed player (fdfc658)
• docs: improved getting started, tool docs, and docstrings (5c827eb)
• docs: updated docs for running with ollama (litellm no longer needed) (f258602)
• docs: updated docs for running locally with ollama/litellm (fa59310)
• improve: enhance API key setup UX and error handling (7231aa9)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.19.2...v0.20.0

These are the release notes for gptme version v0.19.2.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.19.1:

📦 gptme

🐛 Fixes (3)

• fix: catch exception if history file fails to load for any reason (d832c9c)
• fix: change openrouter default model to llama-3.1-405b-instruct (fcf69f5)
• fix: openrouter wasn't picked up as default provider when apikey present (fixes #157) (a8dc5fd)

🔨 Misc (5)

• chore: bump version to 0.19.2 (cea928e)
• refactor: extracted chat and step from cli.py into new chat.py (#158) (5081c89)
• docs(README): change pypi/pepy badges/links to use new gptme package (1a7e4ef)
• docs(README): fix pepy.tech badge links, until the new package name doesn't 404 (8f7d272)
• docs(README): fixed logo in PyPI readme by using full URL (cced9cc)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.19.1...v0.19.2

These are the release notes for gptme version v0.19.1.

Contributors

Thanks to everyone who contributed to this release:

@cshintov, @erikbjare

Changelog

Changes since v0.19.0:

📦 gptme

✨ Features (1)

• feat: added diff previews for save tool (9694fad)

🐛 Fixes (4)

• fix: fixed typing (21c550c)
• fix: hopefully fix flaky test (8a99808)
• fix: shorten output paths by replacing home path with tilde (5d66fda)
• fix: fixed vision issues for .jpg files and when resuming convo (548634d)

🔨 Misc (14)

Click to expand

• chore: bump version to 0.19.1 (eb1c17f)
• chore: changed PyPI name from gptme-python to gptme (fixes #148) (354d0c2)
• refactor: moved ./static into ./gptme/server/static (b47a7bc)
• docs: add local provider in provider examples (#149) (f54514b)
• refactor: moved ./train into ./scripts (cab7a76)
• docs: move CONTRIBUTING.md to page in docs (ea4f30a)
• docs: moved timeline links in README into docs (69b2dea)
• docs(README): center align intro text (945b5ce)
• docs(README): improved intro text (50a45d1)
• docs: fixed relative link in README (broken on site) (174eda5)
• docs: fixed link replacement in site generation (afbb754)
• docs: fixed the site style a bit (6339e2d)
• docs: improved site generation from README.md through better use of pandoc styling and templating (2c41b63)
• docs(README): updated vision as complete (c3330ac)

(excluded 3 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.19.0...v0.19.1

These are the release notes for gptme version v0.19.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.18.2:

📦 gptme

✨ Features (3)

• feat: added screenshot tool (#92) (f4c63c2)
• feat: added vision tool (597c66c)
• feat: added -t/--tools option to specify which tools to load (48d559b)

🐛 Fixes (13)

Click to expand

• fix: fixed lint (a3bcf49)
• fix: removed incorrect calls to register_function, removed dead code (a4fa62e)
• fix: more patch tool refactor (9627b73)
• fix: removed emoji from ask_execute, added secret auto answer in ask_execute (5eddc6a)
• fix: added patch previews (dd7be0d)
• fix: completed diff_minimal (df9d83c)
• fix: flush stdin before asking to execute (prevent unread input from answering before asked) (6ab4dbe)
• fix: implemented example_to_xml to support xml-ify prompt (#146) (5f37d10)
• fix: init tools in evals (44e76ff)
• fix: improved patch warning message on large patches (5e635b6)
• fix: froze more dataclasses (953f801)
• fix: improve tool init logic for tools needing it (python), added toolspec args docs (f8e5cd6)
• fix: use prompt chaining in subagent task to improve reliability (0dd6583)

🔨 Misc (8)

Click to expand

• chore: bump version to 0.19.0 (bbe3586)
• docs: remove completed TODOs (337b025)
• refactor: refactor patch tool, with plans for producing minimal diffs to replace inefficient diffs in log (e2b2d6a)
• docs: fixed docstring for gptme.prompts (027cb06)
• docs: keep copyright year up-to-date (1ac4aee)
• docs: fixed automation docs (a70b5b5)
• docs: changed erik.bjareholt.com/gptme/ links to gptme.org/ (52baf5b)
• docs: added minimal automation docs (#144) (3902b86)

(excluded 2 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.18.2...v0.19.0

These are the release notes for gptme version v0.18.2.

Contributors

Thanks to everyone who contributed to this release:

@brayo-pip, @erikbjare

Changelog

Changes since v0.18.1:

📦 gptme

✨ Features (1)

• feat: major improvments to evals, can now aggregate eval_results and run rust+browser+git in Docker (passes evals) (bfd1b25)

🐛 Fixes (13)

Click to expand

• fix(nit): removed excessive newlines in eval output (780c8f9)
• fix: improved tabcomplete to work well for paths with '~' (b4ff00e)
• fix: improved code coverage, removed CMDFIX constant (5a7c25a)
• fix: fixed typing and remove unused function (9c88947)
• fix(eval): remove gen/run stdout/stderr file paths from results csv (9af35be)
• fix(eval): dont use CRLF line endings in csv's (67ae863)
• fix(eval): strip 'openrouter/' prefix when displaying aggregated results (77e9922)
• fix: fixed lint (2461448)
• fix: removed code from CaseResult (9481c91)
• fix: sort models in eval output, change N/A emoji to questionmark (ab63900)
• fix: catch exceptions in shell tool (f42be8f)
• fix: sort eval output (a2814a3)
• fix: made failed patch messages less verbose (3a408d3)

🔨 Misc (5)

• chore: bump version to 0.18.2 (81708e6)
• tests: update models to provider/model format (#141) (9060e15)
• docs: add "live" eval result output to docs (03f8972)
• refactor: moved Dockerfiles into ./scripts (46340ca)
• tests: fixed browser search (when desc element not found) (bad47ef)

(excluded 5 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.18.1...v0.18.2

These are the release notes for gptme version v0.18.1.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.18.0:

📦 gptme

🐛 Fixes (1)

• fix: expanduser() in patch tool, output warning suggesting save tool for inefficient patches (ca2d4e5)

🔨 Misc (1)

• chore: bump version to 0.18.1 (b0c2ae3)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.18.0...v0.18.1

These are the release notes for gptme version v0.18.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.17.0:

📦 gptme

✨ Features (1)

• feat: added better interrupt handling, requiring two interrupts within 1 sec to exit (when not generating response or executing tools) (7029059)

🐛 Fixes (16)

Click to expand

• fix: nit to python tool instructions (94b1aaf)
• fix: refactored cli params, removed special 'ask' value for name (cbf1d12)
• fix: persist pinned and hide to jsonl, only include files, pinned, etc in toml output if set (c54c1f2)
• fix: fixed bug in refactor (43f912e)
• fix: fix resume after refactor (5d1a761)
• fix: minor fixes, set Console.log_path=False, undo /exit message before exit (711cab4)
• fix: fixes to cli, improved interrupt, refactored conversation picking to not run when piped, dont run assistant until user message present (project context fix) (417b319)
• fix: improve rich usage, change calls to use gptme.util.console.{print,input,log} (8cf53cb)
• fix: improved browser tool search output, if python tool had result then skip stdout in msg (4aaf202)
• fix: updated system prompt to mention tags (c686dab)
• fix: limit shell output (8a62859)
• fix: limited default number of listed conversations to 20 in webui (84ab220)
• fix: added OpenRouter url when asking for API key (8728012)
• fix: fix conversation list order in picker, lazily load conversation metadata, add get_user_conversations(), add ?limit= to /api/conversations and use it in webui (9c53aa0)
• fix: set gptme.version (abcfec0)
• fix: fixed prompt chaining, added test (fixes #106) (deac8db)

🔨 Misc (15)

Click to expand

• chore: bump version to 0.18.0 (4a7404f)
• refactor: separated cli/interactive stuff from chat() into main() (9808166)
• docs: added more examples (da88a5f)
• docs: improved docs config, fixed warnings, build in strict mode in CI (no warning allowed) (ae45141)
• tests: fixed browser search test (a106d5c)
• docs(README): added ToC (e88d426)
• docs: added link to examples from intro (6d1471e)
• refactor: renamed function to remove 'private' underscore prefix (a7a5cf6)
• docs: updated README (5983f5b)
• tests: fixed test (d0a946b)
• docs: added TODO comment (8ad35e1)
• docs: minor improved examples (1158d95)
• docs: improved docs structure (User & Dev guide), improved CLI & API Reference, extracted Prompts as new page, added Examples (e1b881a)
• format: fixed formatting and typing (68c2552)
• docs: added external link to docs index/sidebar (c9bc488)

(excluded 8 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.17.0...v0.18.0

These are the release notes for gptme version v0.17.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.16.0:

📦 gptme

✨ Features (1)

• feat: add support for XML-formatted tool calls (#121) (d0b070f)

🐛 Fixes (23)

Click to expand

• fix: changed warning log to debug (0c4c186)
• fix: minor eval refactor (improved type names), clarified python tool instructions (e0c79a4)
• fix: disable tqdm in tests (77ef4fa)
• fix: reduced number of decimals in eval output (b216179)
• fix: removed spammy message when not in a git repo (be3d023)
• fix: futher reliability improvements to evals (622e574)
• fix: added tqdm progress bar to eval (cd33e06)
• fix: nitpick (11f1e7e)
• fix: improved typing in gptme.evals.run (6d00be7)
• fix: more fixes, speed up list_chats and search_chats by lazily searching chronologically (8c3cb77)
• fix: more fixes and store eval case results in result directory (1e23ecb)
• fix: comment out warning for unknown codeblock types, add lru_cache to frequently called get_tool_for_langtag, added wip llm_openai.list_models (aff213f)
• fix: refactored evals, fixed leaked semaphore warnings, read logs and naively compute tokens from output (3bbd88b)
• fix: improved system prompt, added system prompt to docs (#123) (62220b1)
• fix: refactored and improved evals (#122) (e4eb81c)
• fix: improved eval stream capturing logic (eecabac)
• fix: improved evals output capturing, don't capture by default if a single test is run, and write streams to results directory (8e9ad4c)
• fix: process eval run futures in the order they are finished, instead of waiting in order (5601fae)
• fix: remove the missing datascience packages warning at python tool init (09f115f)
• fix: add ignorelist for certain known non-executable codeblock langs (7b82b03)
• fix: lowered logging level for some spammy messages (9a729af)
• fix: disable placeholder-aware patching if placeholders in original file, improve error message if file not found (1a1e9fb)
• fix: dont ask for version in bump_version script if already on tag and pyproject updated (68aae66)

🔨 Misc (11)

Click to expand

• chore: bump version to 0.17.0 (3044149)
• tests: fixed test_search_chats test (56ffb5a)
• docs: updated README (a837b32)
• tests: added particle effect integration-test example (5aa7955)
• tests: fixed broken test in CI (4947265)
• refactor: refactored Message into a frozen dataclass (8887ca8)
• refactor: moved openai model metadata into seperate file, added make update-models to use gptme to update it (11221db)
• chore: updated gitignore (f2b8be1)
• docs: updated README (ee87599)
• tests: fixed test_eval_cli running on models than intended, and not the tested provider (d0ab042)
• refactor: refactored eval TypedDict types to dataclasses (25de7f7)

(excluded 5 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.16.0...v0.17.0

These are the release notes for gptme version v0.16.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.15.0:

📦 gptme

✨ Features (7)

• feat: basic support for openai/o1-preview and openai/o1-mini (#117) (cf13bae)
• feat: added youtube tool (#116) (ad66992)
• feat: support placeholders in patches (#114) (2e08a3a)
• feat: added read_chat function to chats tool (#115) (527ce5a)
• feat: added list_chats to chats tool, and cleaned up/refactored non-ToolSpec-using tools (#110) (5cb3936)
• feat: added a basic tool to search past conversation logs (#109) (4e36109)
• feat: added terminal bell to alert the user they have been returned control (1da7d04)

🐛 Fixes (13)

Click to expand

• fix: support multiple patches in a single codeblock (#118) (ae3ea89)
• fix: made eval harness more reliable, using Manager (#119) (0787f59)
• fix: remove spammy log message when youtube tool not available (6ab895c)
• fix: extended the patch tool prompt to not strictly forbid placeholders, mention scoping strategies and fallback to save (7cf2119)
• fix: added ./projects and ./demos to gitignore (ef7d9fd)
• fix: change error on unknown codeblock langtags into warning, dont warn on empty langtag (481ab38)
• fix: minor improvements to --help output, updated --help example output in README (78f461c)
• fix: fixed excessive whitespace in patch example prompt (f0818c3)
• fix: fixed formatting in tools/base.py (e3987bd)
• fix: added gptme/server/init.py (02a3ecf)
• fix: improved how ToolUse examples are formatted (1e6574f)
• fix: fixed bug in how examples were generated for patch tool (c8489ec)
• fix: dont crash on unknown shell syntax (dd6fff7)

🔨 Misc (5)

• chore: bump version to 0.16.0 (d7f170c)
• tests: bumped short token allowed token length (0208e76)
• refactor: refactor tools, codeblock, and tooluse (#113) (0cad5ca)
• tests: added minimal tests for chats tool (d079af4)
• format: formatted codebase (740f132)

(excluded 3 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.15.0...v0.16.0

These are the release notes for gptme version v0.15.0.

Contributors

Thanks to everyone who contributed to this release:

@erikbjare

Changelog

Changes since v0.14.2:

📦 gptme

✨ Features (4)

• feat: added screenshot_url function to browser tool (9cd3850)
• feat(bot): support non-change questions/answers (de9844d)
• feat: added special prompting for --non-interactive (9d912c3)
• feat(github-action): use Docker for gptme execution (97f8724)

🐛 Fixes (11)

Click to expand

• fix: minor improvement to patch tool prompt (722b2c4)
• fix: fixed bug with extra 'Skipped hidden system messages' lines from workspace prompt (5937cb7)
• fix(anthropic): fixed vision and other issues with preparing messages (b9b8455)
• fix: don't include paths for slash-command arguments, dont inlude workspace prompt on resume (4900d19)
• fix: catch rich.print() errors and fall back to builtins.print() when printing messages (8698bef)
• fix: fixed bug checking for browser tool when not available (f86569d)
• fix: set session size for tmux tool, fixed wrong tmux examples (terminal instead of tmux) (1ef4587)
• fix: changed tabulate tablefmt in eval output (4251854)
• fix: fixed docker workspace permissions in bot action (aee7f95)
• fix: switch from timeout-minutes to using timeout command in bot action step (fa23c66)
• fix: add 'shell' to shell tool's block_types (d2c4879)

🔨 Misc (9)

Click to expand

• chore: bump version to 0.15.0 (01c4812)
• docs: added TODO comment in python tool about which venv the repl should ideally run in (0d8eb85)
• docs: fixed bad reference (780b3c8)
• refactor(eval): refactored gptme.eval module, splitting gptme.eval.evals into gptme.eval.suites.{basic, init-project, browser} (bfe5e1b)
• docs: removed Inputs section from bot.md (e18dcb1)
• docs: renamed webui.rst to server.rst, added bot.md to index and updated instructions (18b7268)
• docs: added example of eval run output (c1987fa)
• tests: fixed blinking test (c6953b3)
• tests: clarified that we're testing so that it doesnt try to show runnable examples (8514bd3)

(excluded 20 less relevant commits)

Full Changelog: https://github.com/ErikBjare/gptme/compare/v0.14.2...v0.15.0