LangChain: Fireworks Caching and Fallback Safety

Fireworks prompt-cache routing and fallback sanitization shipped in langchain-fireworks 1.7.0 and langchain 1.4.3. The rest of the activity hardens streaming, new-model compatibility, and core tool execution.

Duration: PT8S

Episode overview

This episode is a short developer briefing from LangChain.

It explains recent repository work in plain language.

  • Show: LangChain
  • Published: 2026-09-29T13:02:28Z
  • Audio duration: PT8S

Transcript excerpt

This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.

Good morning, it's Tuesday, September 29th, 2026, and this is your LangChain briefing.

The lead pattern is cheaper, more resilient multi-model agents, centered on Fireworks prompt-cache reuse that survives fallbacks.

First, cache affinity with safe fallback. Pull request 38823 adds Fireworks prompt caching middleware that derives session affinity from a hash of the thread ID, keeping calls in the same agent thread on the same replica. Explicit user, cache key, and affinity settings take precedence, and nothing is generated…

Second, streaming and new-model compatibility fixes. For Anthropic, pull request 40882 adds Claude Sonnet 5.5 profiles and handling for thinking settings, forced tool choice, and system changes, while pull request 40893 preserves per-message output config and turn-scoped system boundaries. For OpenAI, pull request…

What's next: if you run Fireworks agents with fallbacks, pick up the new releases and verify cache settings and timeout handling after output starts. If you target Sonnet 5.5 or OpenAI reasoning models, test structured output, stop parameters, and streaming reconciliation.

That's the briefing for today, thanks for listening.

Nearby episodes from LangChain

  1. Anthropic Replay Fixes
  2. Weekly Recap - Hardening Tool Calls and Provider Catch-Up
  3. Agent Recovery and Tool Guardrails
  4. Streaming Performance and Bedrock Compatibility
  5. Provider Guardrails and Tracing Fixes
  6. Mid-Conversation Tool Changes and Typesafe Hardening
  7. Tool Calling Hardening and Opus 5.5 Readiness
  8. Tool Reliability Fixes