Recall that actually works
Hybrid semantic + keyword search returns only what's relevant to the task in front of you — not a pile of noise.
~4 ms recall · p95
Local-first · $0 · Offline · Open source
Persistent memory for your AI coding agent. It remembers every session — so you stop re-explaining the same context, every day. Local, free, private.
Works on macOS, Windows & Linux. Node ≥ 22.
Works with your agent — five harnesses, one shared memory
Why it exists
Every session, it starts from zero. agentbrainsystem remembers — captures each one, recalls what matters, hands-free.
Hybrid semantic + keyword search returns only what's relevant to the task in front of you — not a pile of noise.
~4 ms recall · p95
Captures each session when it ends and feeds the right context back at the next start. One command — then it's invisible.
$0 · no LLM · hands-free
100% local and offline. No cloud, no account, no API keys required, no telemetry. The model downloads once, then runs offline.
offline · private · $0
How it works
Hooks (or an OpenCode plugin) auto-ingest every coding session when it ends — $0, no LLM.
JSONL · hooks
Embedded locally on your machine, fully offline. Nothing is uploaded.
SQLite · sqlite-vec · FTS5
Hybrid semantic + keyword search surfaces relevant memory — at session start and on every prompt, as you work.
per-prompt · MCP recall
More than recall
abs ui
Your whole store rendered as one bioluminescent jellyfish: the dome is the consolidated core, each tentacle a session, each bead of light an observation. Zoom, orbit, inspect, prune.
FAQ
abs setup offers a guided, skippable LLM step that sharpens recall — local Ollama is $0 and offline, or any hosted OpenAI-compatible endpoint; your key is never stored. Skip it and recall still works.