l33tdawg/sage

v7.1.0 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 17d AI Agents & Assistants

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agentic-ai artificial-intelligence bft-consensus bft-network distributed-systems governance

Summary

AI summary

Added query-expansion support on POST /v1/memory/hybrid.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Medium	Added --expand N flag to bench(longmemeval) for LLM-generated query variants (v7.1) Added --expand N flag to bench(longmemeval) for LLM-generated query variants (v7.1) Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Added v7.1 LongMemEval-S + LoCoMo result JSONs to bench(results) Added v7.1 LongMemEval-S + LoCoMo result JSONs to bench(results) Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Integrated LoCoMo seed-once, Makefile fetch, sidecar hygiene, and LongMemEval LLM-judge into bench Integrated LoCoMo seed-once, Makefile fetch, sidecar hygiene, and LongMemEval LLM-judge into bench Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Added native reranker sidecar with backend disclosure in result JSON (v7.1) Added native reranker sidecar with backend disclosure in result JSON (v7.1) Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Implemented LoCoMo retrieval-benchmark harness (v7.1) Implemented LoCoMo retrieval-benchmark harness (v7.1) Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Added query-expansion support on POST /v1/memory/hybrid (v7.1) Added query-expansion support on POST /v1/memory/hybrid (v7.1) Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Enabled node-operator read-scope bypass for SessionStart prefetch via hooks (v7.1) Enabled node-operator read-scope bypass for SessionStart prefetch via hooks (v7.1) Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Added hybrid recall support on Python SDK with v7.1.0 Added hybrid recall support on Python SDK with v7.1.0 Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Feature	Medium	Implemented cross-encoder reranker on hybrid recall path (v7.1) Implemented cross-encoder reranker on hybrid recall path (v7.1) Source: granite4.1:8b-q6_K@2026-05-21 Confidence: high	—
Dependency	Medium	Baked PYTHONUNBUFFERED=1 into Makefile targets for bench Baked PYTHONUNBUFFERED=1 into Makefile targets for bench Source: granite4.1:8b-q6_K@2026-05-21 Confidence: low	—
Bugfix	Medium	Tidied reranker response struct conversion in embedding Tidied reranker response struct conversion in embedding Source: granite4.1:8b-q6_K@2026-05-21 Confidence: low	—
Bugfix	Medium	Used ${CLAUDE_PROJECT_DIR} for hook script paths in hooks Used ${CLAUDE_PROJECT_DIR} for hook script paths in hooks Source: granite4.1:8b-q6_K@2026-05-21 Confidence: low	—
Other	Medium	Removed 'built on real research' info-note from site docs, kept v7.0 benchmark callout Removed 'built on real research' info-note from site docs, kept v7.0 benchmark callout Source: granite4.1:8b-q6_K@2026-05-21 Confidence: low	—
Other	Medium	Refreshed landing page for v7.0 in site documentation Refreshed landing page for v7.0 in site documentation Source: granite4.1:8b-q6_K@2026-05-21 Confidence: low	—

Full changelog

SAGE v7.1.0

Sovereign Agent Governed Experience — persistent, governed memory for AI agents.

Quick Start (sage-gui)

# Download and extract for your platform, then:
./sage-gui setup    # Interactive setup wizard
./sage-gui serve    # Start your personal memory node

See the README for full documentation.

Changelog

c82f4d1c6161e4b66867d6165a235506930b6a0c bench(longmemeval): add --expand N flag for LLM-generated query variants (v7.1)
be6bb9a5c8bb05585220635c823a5675c35b549f bench(longmemeval): bake PYTHONUNBUFFERED=1 into Makefile targets
ca0c73faf32f05becb47bf1c16020ca8da711639 bench(results): v7.1 LongMemEval-S + LoCoMo result JSONs
2052514b9e227050d5d3c7cc267a4d6c49431d08 bench: LoCoMo seed-once + Makefile fetch + sidecar hygiene + LongMemEval LLM-judge
733be45b9430c6cc7b40662ad8228a93eaa32a3b bench: native reranker sidecar + backend disclosure in result JSON (v7.1)
765dc91b71e9a3703f86e32aa168e35cd3a7e89b docs(site): drop 'built on real research' info-note, keep only the v7.0 benchmark callout
dcb6b1efc7e2e962e77cc2e4f58bf085d30e3cf2 docs(site): refresh landing page for v7.0
0d098666535d5f43e047d65a1a24f07f45469d3a feat(bench): LoCoMo retrieval-benchmark harness (v7.1)
6089ba024ef821651d1a367837df3ada578dfc7b feat(rest): query-expansion support on POST /v1/memory/hybrid (v7.1)
36084a3217a7e23231c0de208f1bc779c1c13b40 feat(rest,hooks): node-operator read-scope bypass for SessionStart prefetch (v7.1)
e8f02a4b89e168b312aa579a368259b868b37f68 feat(sdk): hybrid recall on Python SDK + v7.1.0
daf3ce7221feedfb94f607a864c5f779d05ce48b feat(store,embedding): cross-encoder reranker on hybrid recall path (v7.1)
dca649cb544b968b75d70cef2d068a36b1bd7573 fix(embedding): tidy reranker response struct conversion
8d77afea1d13e9accbe027441d195656b12b9cc3 fix(hooks): use ${CLAUDE_PROJECT_DIR} for hook script paths

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 l33tdawg/sage

Get notified when new releases ship.

About l33tdawg/sage

Institutional memory for AI agents with real BFT consensus. 4 application validators vote on every memory before it's committed — no more storing garbage. 13 MCP tools, runs locally, works with any MCP-compatible model. Backed by 4 published research papers.

All releases →