Skip to content

BetterDB-inc/monitor

vagent-cache-v0.7.0 scope: agent-cache Breaking

This release includes 2 breaking changes for platform teams planning a safe upgrade.

Published 1mo Monitoring & Metrics
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

prometheus monitoring obsevability redis valkey

Affected surfaces

breaking_upgrade

ReleasePort's take

Light signal
editorial:auto 1mo

The Vercel AI SDK and LlamaIndex adapters now incorporate tool definitions into cache keys, causing existing entries to miss once. The LlamaIndex `prepareParams()` function also gains an optional `tools` parameter.

Why it matters: Cache key changes trigger one‑time misses for all prior cached tool‑using requests; plan for warm‑up after upgrade. Severity rated 70 indicates high impact on cache hit rates.

Summary

AI summary

Cache keys now incorporate tool definitions for Vercel AI SDK and LlamaIndex adapters, causing prior cached entries to miss once.

Changes in this release

Breaking High

Cache keys for tool-using requests changed on Vercel and LlamaIndex adapters; existing entries become one-time misses.

Cache keys for tool-using requests changed on Vercel and LlamaIndex adapters; existing entries become one-time misses.

Source: llm_adapter@2026-06-11

Confidence: high

 —
Feature Low

LlamaIndex `prepareParams()` now accepts an optional `tools` parameter, defaulting to prior messages‑only behavior if omitted.

LlamaIndex `prepareParams()` now accepts an optional `tools` parameter, defaulting to prior messages‑only behavior if omitted.

Source: llm_adapter@2026-06-11

Confidence: high

 —
Bugfix Medium

Vercel AI SDK adapter includes tool definitions in cache key.

Vercel AI SDK adapter includes tool definitions in cache key.

Source: llm_adapter@2026-06-11

Confidence: high

 —
Bugfix Medium

LlamaIndex adapter includes tool definitions in cache key when `tools` is passed.

LlamaIndex adapter includes tool definitions in cache key when `tools` is passed.

Source: llm_adapter@2026-06-11

Confidence: high

 —
Full changelog

Fixed

  • Vercel AI SDK adapter: tool definitions now included in cache key. Requests with identical messages but different tools no longer share cached responses. seed, stopSequences, responseFormat, and toolChoice are also part of the key now.
  • LlamaIndex adapter: tool definitions now included in cache key when tools is passed to prepareParams(). Only serializable metadata is used; the call closure is never serialized.

Changed

  • Cache keys changed for tool-using requests on Vercel and LlamaIndex adapters. Existing cached entries for those requests will be a one-time miss after upgrade — intended, as the prior entries were not safe to reuse across differing tool sets.
  • LlamaIndex prepareParams() now accepts a tools option. Omitting it falls back to messages-only keying (prior behavior).

Known limitations

  • LangChain adapter: tool-schema drift is not reflected in the cache key — the framework's BaseCache interface only exposes (prompt, llm_string). Documented limitation, unchanged.

Full Changelog: https://github.com/BetterDB-inc/monitor/compare/agent-cache-v0.6.1...agent-cache-v0.7.0

Full Changelog: https://github.com/BetterDB-inc/monitor/compare/v0.23.0...agent-cache-v0.7.0

Breaking Changes

  • Cache keys for tool-using requests changed on Vercel AI SDK and LlamaIndex adapters; existing cached entries will be invalidated once (one‑time miss).
  • `prepareParams()` in LlamaIndex now accepts a `tools` option, defaulting to messages‑only keying when omitted.
View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track BetterDB-inc/monitor

Get notified when new releases ship.

Sign up free

About BetterDB-inc/monitor

Valkey-first observability with Redis compatibility. Query real-time metrics, analyze slow commands, detect hot keys, and investigate performance issues directly from AI coding assistants.

All releases →

Related context

Earlier breaking changes

  • vagent-cache-py-v0.7.0 Cache keys changed for tool-using requests on the LlamaIndex adapter; existing cached entries will miss once after upgrade.
  • vagent-v1.5.0 TLS is now mandatory; agent exits with error if `--valkey-tls true` is not set for IAM auth mode.

Beta — feedback welcome: [email protected]