Skip to content

sgx-labs/statelessagent

v0.6.0 Security

This release includes 4 security fixes for security teams reviewing exposed deployments.

Published 3mo MCP Developer Tools
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →
This release patches 4 known CVEs

Topics

ai-agent ai-memory claude-code context-surfacing cursor go
+12 more
llm-memory local-first mcp mcp-server model-context-protocol obsidian llm semantic-search sqlite sqlite-vec vector-db windsurf

Affected surfaces

auth rbac

Summary

AI summary

Write‑side MCP tools add note, decision and handoff creation capabilities.

Full changelog

v0.6.0 — Reliability, Privacy & Polish

Self-diagnosing retrieval, pinned notes, keyword fallback, vault privacy structure, RAG chat, interactive demo, write-side MCP tools, security hardening, and a full polish pass.

Install

curl -fsSL statelessagent.com/install.sh | bash
same demo

Or download a binary below.


Added

  • Write-side MCP tools — 5 new MCP tools bring the total to 11. Your AI can now save notes, log decisions, and create session handoffs — not just read:
    • save_note — create or update markdown notes (auto-indexed, dot-dir protected, 100KB limit)
    • save_decision — log structured decisions with status and date
    • create_handoff — session handoffs with summary, pending items, and blockers
    • get_session_context — one-call orientation: pinned notes + latest handoff + recent activity + stats
    • recent_activity — recently modified notes (clamped to 50)
  • same ask — ask questions, get answers FROM your notes with source citations. Uses a local Ollama LLM to synthesize answers from semantically relevant notes. 100% local, no cloud APIs.
  • same demo — interactive demo with sample notes. See SAME in action in under 60 seconds. Works without Ollama.
  • same tutorial — 6 hands-on lessons: search, decisions, pinning, privacy, RAG, and session handoffs.
  • SAME Lite — works without Ollama. Keyword search via SQLite FTS5 powers all features. Install Ollama later and same reindex upgrades to full semantic mode.
  • Project-aware init — detects existing docs (README.md, docs/, ARCHITECTURE.md, CLAUDE.md, .cursorrules, ADR/) and indexes them automatically.
  • same pin — pin important notes so they're always included in every session.
  • same repair — one-command database recovery: backs up and force-rebuilds.
  • same feedback — thumbs-up/down for notes to improve retrieval over time.
  • FTS5 keyword fallback — when Ollama is down, context surfacing falls back to keyword search instead of failing.
  • Doctor retrieval diagnostics — 8 new checks: embedding mismatch, PRAGMA integrity, utilization rate, config validity, and more.
  • Schema migration system — auto-migrates between schema versions.
  • Embedding mismatch guard — detects provider/model changes without reindexing.
  • MCP directory manifestsserver.json and smithery.yaml for directory submissions.
  • 45+ new tests — store, search, indexer, config, and MCP packages.

Security

11 fixes from 6 rounds of pre-release security auditing:

  • Dot-path blocking — MCP writes can no longer touch .git/, .same/, .env, etc.
  • MCP error sanitization — all error messages are static strings; no internal paths leak to AI
  • find_similar_notes path validation through safeVaultPath
  • Write size limits — 100KB max on decisions and handoffs
  • Config file permissions — 0o644 → 0o600 across all config writes
  • OLLAMA_URL scheme validation — blocks file://, ftp://
  • Empty input validation on same search, same ask, same feedback
  • DB path PII fix — index_stats returns filename only, not full path
  • Plugin timeout safety — nil check before Kill()

Fixed

  • Replaced all panics with errors
  • TOML skip_dirs now applied correctly
  • Noise path filter uses HasPrefix instead of Contains

Changed

  • Go 1.25 standardized everywhere
  • README restructured for launch — same demo above the fold, MCP tools table, numbers, SAME Lite callout
  • Intel Mac install — uses ARM binary + Rosetta
  • Error messages rewritten for clarity

Full Changelog: https://github.com/sgx-labs/statelessagent/compare/v0.5.4...v0.6.0

Security Fixes

  • Dot-path blocking prevents MCP writes to sensitive directories (.git/, .same/, .env, etc.)
  • MCP error sanitization ensures all error messages are static strings with no internal path leakage
  • Input validation fixes for `find_similar_notes`, empty queries in `same search/ask/feedback`, and OLLAMA_URL scheme restrictions (block file://, ftp://)
  • Write size limits enforce a 100 KB maximum on decisions and handoffs

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

Track sgx-labs/statelessagent

Get notified when new releases ship.

Sign up free

About sgx-labs/statelessagent

Memory with provenance tracking — records where agent knowledge originated and detects when sources change. 17 MCP tools for session handoffs, decisions, semantic search, and knowledge graph. Works across Claude Code, Cursor, Windsurf, Codex CLI, and Gemini CLI. Single Go binary, SQLite + vector search, fully local.

All releases →

Beta — feedback welcome: [email protected]