This release adds 4 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+6 more
Summary
AI summaryPrompt and response caching across cloud providers cuts multi-turn agent costs by up to 80%.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Adds prompt caching across 7 cloud providers reducing multi-turn costs by 50‑80% automatically. Adds prompt caching across 7 cloud providers reducing multi-turn costs by 50‑80% automatically. Source: llm_adapter@2026-05-25 Confidence: high |
— |
| Feature | Medium |
Adds response cache with exact and optional semantic match, opt‑in per step. Adds response cache with exact and optional semantic match, opt‑in per step. Source: llm_adapter@2026-05-25 Confidence: high |
— |
| Feature | Medium |
Enables auto-compaction by default with threshold lowered to 80K chars, fully backwards‑compatible. Enables auto-compaction by default with threshold lowered to 80K chars, fully backwards‑compatible. Source: llm_adapter@2026-05-25 Confidence: high |
— |
| Feature | Medium |
Adds tool memoization for deterministic tools (code_search, pdf_parser, etc.). Adds tool memoization for deterministic tools (code_search, pdf_parser, etc.). Source: llm_adapter@2026-05-25 Confidence: low |
— |
| Feature | Medium |
Adds Cache Analytics dashboard showing $ saved, hit rates, and per‑step badges. Adds Cache Analytics dashboard showing $ saved, hit rates, and per‑step badges. Source: llm_adapter@2026-05-25 Confidence: low |
— |
| Feature | Low |
Implements tool memoization for deterministic tools (code_search, pdf_parser, xlsx_parser, time, code_indexer, collect_data, personal_details). Implements tool memoization for deterministic tools (code_search, pdf_parser, xlsx_parser, time, code_indexer, collect_data, personal_details). Source: granite4.1:30b@2026-05-25-audit Confidence: low |
— |
| Feature | Low |
Provides Cache Analytics dashboard showing total dollars saved, per‑model hit rates, and top‑run breakdowns. Provides Cache Analytics dashboard showing total dollars saved, per‑model hit rates, and top‑run breakdowns. Source: granite4.1:30b@2026-05-25-audit Confidence: low |
— |
Full changelog
Major Release - Cache Layer
- Prompt caching wired into all 7 cloud providers (Anthropic cache_control, OpenAI/Grok/DeepSeek/Gemini auto-cache extraction, Bedrock cachePoint). Cuts multi-turn agent costs by 50–80% automatically.
- Response cache (exact + optional semantic match) — opt-in per step on LLM/Evaluator/Extract-JSON nodes. Cache hits skip the LLM call entirely.
- Tool memoization for deterministic tools (code_search, pdf_parser, xlsx_parser, time, code_indexer, collect_data, personal_details). Side-effectful tools (bash, sql_agent, web_scraper, sandbox) are never cached.
- Cache Analytics dashboard in Usage tab — total $ saved, per-model hit rates, top-run breakdowns. Per-step cache: hits/total badges on the canvas.
- Auto-compaction enabled by default (auto_compact_threshold lowered to 80K chars). Fully backwards-compatible — existing orchestrations/settings/usage logs upgrade with zero migration.
What's Changed
- Feature/cache mechanism by @naveenraj-17 in https://github.com/synapseorch-ai/synapse-ai/pull/303
Full Changelog: https://github.com/synapseorch-ai/synapse-ai/compare/v1.5.6...v1.6.0
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
About synapseorch-ai/synapse-ai](https:
All releases →Related context
Related tools
Beta — feedback welcome: [email protected]