v0.14.0

v0.14.0#

These are the release notes for gptme version v0.14.0.

Contributors#

Thanks to everyone who contributed to this release:

@erikbjare

Changelog#

Changes since v0.13.2:

📦 gptme#

✨ Features (2)#

  • feat: anthropic prompt caching beta (bb3d9a58)

  • feat: started working on vision (7b0c5b0f)

🐛 Fixes (25)#

Click to expand

  • fix: improved test flakiness (0758c4d4)

  • fix: check if gitignore exists (d664f4b2)

  • fix: moved vue@create example into bash tool (dont try to do it interactively) (5ed36eb8)

  • fix: switch to ```ipython syntax for executing with python tool, to differentiate from code samples (see #67) (42ee56ab)

  • fix: changed save format to be more explicit (68077b6d)

  • fix: removed spammy warning (90aa4a62)

  • fix: removed use of NotRequired for TypedDict (not in Python 3.10) (cfce1303)

  • fix: use anthropic api max_retries instead of custom retry_anthropic decorator (008f58c5)

  • fix: added retry_anthropic decorator for rate limits (d37d7fbc)

  • fix: fixed to evals, capture eval output on timeout/terminate (cd0862a7)

  • fix: fixed spammy prints (855a46b0)

  • fix: clarified return format for subagent (e5e2a9ab)

  • fix: fixed bug in transform_examples_to_chat_directives (2997aa43)

  • fix: print logs dir with –version command (1ca6127f)

  • fix: enabled stricter linting and fixed lints (apparently needed given a25aa7d369dec79341500bcf735e4237def53052) (bf67b323)

  • fix: fixed nasty bug with mutable argument default (a25aa7d3)

  • fix: switch recommended openai model to gpt-4o (b3582aca)

  • fix: moved len_tokens and msgs2dicts from util.py to message.py (94bade33)

  • fix: support nested codeblocks, rewrote/refactored codeblock parsing/management (3e291a4f)

  • fix: fixed incorrect storage format for Message.to_dict (07f1cbbf)

  • fix: added preliminary nested codeblock support (885e544b)

  • fix: added openai vision support (6bbec93b)

  • fix: completed basic vision support (f1846079)

  • fix: correct extremely short ‘Thinking…’ message, now wait until first character (b25e576c)

  • fix: typing for subagent tool, added –check-untypes-defs to mypy (d94a71e5)

🔨 Misc (20)#

Click to expand

  • chore: bumped version to v0.14.0 (d4c55b61)

  • refactor: renamed terminal tool to tmux (79355bc6)

  • tests: increase max tokens for full system prompt significantly (c785cfef)

  • tests: made test_subprocess less flaky (34234742)

  • tests: added difficult but reasonably simple integration test example (create vite project, build simple app) (57b6febb)

  • docs(README): minor improvements (c43cc8a1)

  • docs(README): misc minor improvements (c705c801)

  • tests: removed uninteresting examples from test-integration.sh (18c1afa0)

  • docs: added links to top of README (b15d6865)

  • docs(README): added mention that vision works, ‘GPTMe’ -> ‘gptme’ for title, mention OpenRouter support (4e39f634)

  • tests: run both eval tests (cli and direct call) for better coverage somehow (859efbec)

  • tests: fixed tests failing due to new file (706f13c9)

  • refactor: refactored provider-specific code into new files llm_openai.py and llm_anthropic.py (eec82155)

  • tests: fix coverage for eval tests (multiprocessing workaround) (9ef1ec46)

  • tests: added test_eval_cli and cleaned up eval code (c0e5c874)

  • tests: basic test for evals (770ed2ca)

  • tests: fix –version test (4ca5078a)

  • tests: remove ambiguity in subagent fib test prompt (fe063ec1)

  • tests: added test for vision (a740194b)

  • refactor: moved eval code into gptme.eval, added gptme-eval entrypoint, fixed typing (8a1bb097)

(excluded 14 less relevant commits)

Full Changelog: https://github.com/gptme/gptme/compare/v0.13.2…v0.14.0