The open-source project "claude-os" adds persistent memory and knowledge search to Claude Code, Anthropic's coding agent. It lets users store information simply by saying "remember this" in natural language, and automatically restores and searches context across sessions.
Open Source · Claude OS v2.5
Persistent memory for Claude Code — 100% local, free, and controlled in plain English
Just say "Remember this: …" and the memory is auto-recalled in later sessions. A hybrid index of syntactic parsing plus embeddings ends the need to re-explain a project every time — all processing stays on your machine.
~80%
fewer embedding chunks vs. embedding every file
~30s
structural index of a 10,000-file project
100%
local operation · free & open source
Indexing time for a 10,000-file project
Semantic index — embed only the top 20% of files vs. embed everything
~20 min
Claude OS
top 20% only
3–5 hrs
Embed every file
conventional approach
Selective embedding turns hours of indexing into minutes — running in the background.
How a memory persists across sessions
"Remember this: …"
Plain-language input
→
Stored locally
SQLite + sqlite-vec via MCP server
→
Auto-recalled
Retrieved in later sessions
Key specifications
Structural index
tree-sitter · ~30s / 10k files
Semantic index
Top 20% · ~20 min background
Memory backend
SQLite + sqlite-vec · Redis Pub/Sub
MCP server
Port 8051 · Ollama local embeddings
Platform / price
macOS / Linux · free (open source)
v2.5 additions
Cross-KB search · inline health checks
Praised
Significantly reduced token consumption
Context retained across sessions boosts productivity
No more repetitive re-explaining; natural-language control
Watch out for
Initial setup complexity
Indexing time on very large projects
Crowded field of similar memory-extension tools
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…