v0.13.0
These are the release notes for gptme version v0.13.0.
Changelog
Summary:
- Anthropic support (Claude 3.5 Sonnet is really good!)
- New
terminaltool, to run long-running or interactive commands. - New
subagenttool, to call subagents for subtasks (saving on context). - Improved
shelltool, now supports more bash syntax. - Tools & prompting refactor.
- Much improved docs, including tool examples also used in system prompt.
- Improved testing, now more reliable than ever!
Changes since v0.12.0.
📦 gptme
✨ Features (2)
- feat: added tmux terminal tool, major tools refactor (#86) (
0c0f987) - feat: added support for anthropic (#84) (
66fa372)
🐛 Fixes (21)
Click to expand
- fix: fixed bundling static assets + favicon, so that web ui works when installed from pypi (
116f847) - fix: added main.py file to gptme (
1811acd) - fix: improvements to subagent tool (
96d5b86) - fix: added more 'pragma: no cover' and coverage excludes (
2f11864) - fix: improve coverage config (
2eda504) - fix: simplified check for browser tool when fetching URLs in prompt (
d996a50) - fix: fixed search result output format (
eda4f79) - fix: fixed incorrectly formatted code blocks (
9d2907b) - fix: typo (
1e6b440) - fix: fixed broken list_sessions command for terminal tool (
e4492aa) - fix: improved transform_examples_to_chat_directives to be more robust (
b86b3ce) - fix: added support for shell pipeline syntax, added tests (
f0a581d) - fix: made transform_examples_to_chat_directives more reliable (
dc04994) - fix: more installed programs to show in prompt (package managers), mention if platform is macOS in shell tool instructions (
9b9e9ed) - fix: refactor tool-specific prompting logic in prompts.py into tools themselves (
c55d9d1) - fix: fixed evals (
475173e) - fix: made shell tool able to handle function definitions (
893fbfd) - fix: fixed ask for API key on first start, accept and auto-detect anthropic keys (
8b33013) - fix: anthropic fixes and ci, auto-detect provider+model from env (#85) (
8c20800) - fix: improved output of /tokens to include input cost (
714a03a) - fix: added more models' metadata (
59b1f99)
🔨 Misc (14)
Click to expand
- test: added basic extract json tests for terminal tool (
3486590) - chore: bumped version to v0.12.5 (
7b6d5d4) - docs: improved index intro text, added webui docs (
041e8e8) - docs: changed to shields.io pypi badge (
82f9eb1) - refactor: moved server code into gptme.server module (
19fc560) - docs: updated usage in README (
8588bfe) - tests: added test to improve coverage for terminal tool (
156e49a) - docs: updated README (
1d8c1e0) - tests: fixed test for shell tool (
23143bb) - docs: reordered index (
9ba156f) - docs: renamed local-models page to providers (
6781c58) - docs: major docs improvements, transform tool examples to chat directives for docs (
068154e) - tests: widened range for acceptable prompt length (
a63f1c5) - docs(README): added mention of support for anthropic & azure (
dd00d4b)
(excluded 10 less relevant commits)
Full Changelog: v0.12.0...v0.13.0