Repository navigation
What's Changed
- added
cheapaimodelas a named LLM provider: OpenAI-compatible gateway athttps://api.cheapaimodel.ai/v1, keyed byCHEAPAIMODEL_API_KEYwith optionalCHEAPAIMODEL_API_URLoverride, wired intoget_litellm_responseandresolve_model_provider, documented in README and the LLM responses guide (607eb38, 27ef14f, e5b753d) - added Firecrawl as a
search_webprovider via newsearch_firecrawl, which posts to the Firecrawl Search API v2 withFIRECRAWL_API_KEYand maps results to the standard title/link/content dicts (72e0d93, 4977281, e5b753d) - fixed litellm auto tool-call turns to add the final summary completion's token usage to reported usage totals, and to keep token accounting intact when the summary completion returns no usage data (2ceef44, d8e639f, e5b753d)
- streaming litellm responses now snoop raw SSE lines for provider-reported usage/cost and emit a final usage event; serve.py MCP SSE and chat event streams request
include_usage=Trueand report provider-reported cost via newprovider_cost_from_usagehelper (2ceef44) - fixed tool schema generation so array parameters (bare
list,List,Optional[list]) always emit anitemskey, resolving OpenAI rejections of the generated function schemas (5f8a30b) - removed legacy console script entry points from setup.py (
npc-init,npc-claude,npc-codex,npc-gemini,npc-opencode,npc-aider,npc-amp,npc-pluginsince these now should be callable from npcsh (they're not quite yet but that will happen soon) ) (e5b753d)
PRs:
- feat: add cheapaimodel as a named provider by @nicho810 in #335
- Add Firecrawl as a search_web provider by @erikengervall in #336
- fix(gen): include the summary completion in usage for litellm auto tool turns by @pcbeingused333 in #333
- fix to address #338 for openai which rejects any array schema missing… by @cagostino in #339
- cost tracking fixes, cli targets removed by @cagostino in #340
New Contributors
- @nicho810 made their first contribution in #335
- @erikengervall made their first contribution in #336
Full Changelog: v2.1.26...v2.1.27