The Wire · Showcase
LANGCHAIN REFRESHES MODEL PROFILES, OPENROUTER 0.2.7 SHIPS
By RepoJournal · Filed · About AI Agents
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
10 people shipped this
saurya
@saurya
4 cited
dominiccooney
@dominiccooney
2 cited
github-actions[bot]
@github-actions[bot]
1 cited
langchain-oss-model-profiles
@langchain-oss-model-profiles
1 cited
langchain-oss-model-profiles[bot]
@langchain-oss-model-profiles[bot]
1 cited
imnishitha
@imnishitha
1 cited
TheRealSpencer
@TheRealSpencer
1 cited
aikido-autofix[bot]
@aikido-autofix[bot]
1 cited
abeatrix
@abeatrix
1 cited
tofarr
@tofarr
OpenRouter integration lands with updated model profiles including Google's Gemini 3.5 Flash Lite, while core token counting gets a performance fix for multi-tool agents.
LangChain released openrouter 0.2.7 [1] with automated model profile refreshes that add four new models and update 42 existing ones across the OpenRouter provider [2]. The headline addition is Gemini 3.5 Flash Lite, bringing 1,048,576 context tokens to agents that need cheap, long-context reasoning [3]. On the core side, a fix to `count_tokens_approximately` [4] now reuses cached `tool_call_schema` for BaseTool instances instead of recomputing OpenAI schemas on every call. For agents managing dozens of schema-rich tools (particularly those using SummarizationMiddleware), this cuts per-turn token counting overhead significantly. Over at OpenHands, two enterprise key management fixes landed [5] [7]: one stops unnecessary key rotation when LiteLLM returns budget-exceeded or transient errors, and another handles managed profile switches by clearing stale member keys [6] [8]. Cline patched two security vulnerabilities: axios bumped to 1.18.0 [9] and js-yaml to 4.3.0 [10]. The desktop client gained one-time routine support and improved run navigation [11], while VSCode integration fixes now reclaim orphaned fallback terminals [12] and make the Proceed While Running decision sticky across command batches [13].
One email a day. Unsubscribe in one click.
What actually shipped in AI Agents, written up every day — commits, pull requests, releases, and security advisories.
One email a day. Unsubscribe in one click. Read a past issue →
Action items
- → Update langchain-openrouter to 0.2.7 if using OpenRouter for new model coverage langchain-ai/langchain [plan]
- → Audit multi-tool agents using count_tokens_approximately for performance gains with the core token counting fix langchain-ai/langchain [monitor]
- → Patch axios and js-yaml in Cline deployments immediately cline/cline [immediate]
- → If running OpenHands enterprise, verify key rotation behavior is no longer triggered by transient LiteLLM errors OpenHands/OpenHands [plan]
References
- [1] langchain-openrouter==0.2.7 ↗ langchain-ai/langchain
- [2] chore(model-profiles): refresh model profile data ↗ langchain-ai/langchain
- [3] chore(model-profiles): refresh model profile data (#39010) langchain-ai/langchain
- [4] fix(core): use `tool_call_schema` cache for `BaseTool` token counting in `count_tokens_approximately` (#39020) langchain-ai/langchain
- [5] fix(enterprise): avoid rotating keys on LiteLLM non-auth errors ↗ OpenHands/OpenHands
- [6] fix: clear member key on managed profile switch ↗ OpenHands/OpenHands
- [7] fix(enterprise): avoid rotating keys on LiteLLM non-auth errors (#15267) OpenHands/OpenHands
- [8] fix: clear member key on managed profile switch (#15356) OpenHands/OpenHands
- [9] fix(security): bump axios to 1.18.0 to resolve multiple CVEs in axios (#12473) cline/cline
- [10] fix(security): update js-yaml from 4.1.1 to 4.3.0 (#12456) cline/cline
- [11] feat(desktop): add one-time routines and run navigation ↗ cline/cline
- [12] fix(vscode): reclaim unobserved fallback terminals ↗ cline/cline
- [13] fix(vscode): make 'Proceed While Running' sticky to the command batch ↗ cline/cline
