No more re-explaining
Every decision, preference, and fact you share with one agent is instantly available to all of them. Start a new session in any tool — it already knows your stack.
Sigil
A shared memory layer your AI coding assistants never forget.
What is Sigil?
Section titled “What is Sigil?”You use AI coding assistants every day. Claude Code knows your preferences from this morning’s session — but Cursor doesn’t. Kiro learned your architecture last week — but Codex CLI has no idea. Every tool starts cold. Every session re-explains the same context.
Sigil is one shared memory layer that every AI assistant on your machine reads and writes to. When you tell Claude Code that you use Postgres LISTEN/NOTIFY instead of Redis, Cursor knows it too. When you ask Kiro about your event system on Friday, it knows what Claude discovered on Tuesday. The brain travels with you — across tools, across sessions, across machines.
Your data never leaves your control. Memory lives in your own Postgres database. No cloud account, no telemetry, no vendor.
100%Retrieval R@10
33msp50 search latency
~1.5Ktokens per prompt
0config files to edit
Why it matters
Section titled “Why it matters”Works with your tools
Claude Code, Cursor, Kiro, Codex CLI, Continue, Cline, Windsurf. Auto-detected and wired on sigil init. No manual config for the big ones.
Fully local and private
Memory lives in your Postgres. Pick Ollama for embeddings and LLM calls — zero data leaves your machine. Air-gappable by design.
Without Sigil / With Sigil
Section titled “Without Sigil / With Sigil”Without Sigil: You open Claude Code on Tuesday and spend 20 minutes explaining that your team switched from Redis to Postgres for event delivery. On Thursday you open Cursor in the same repo — it has no idea. Friday you’re back in Claude Code, a new session, and it’s gone again.
With Sigil: You mention the Postgres decision once. The Stop hook captures it automatically. On Thursday Cursor retrieves it before you even ask. Friday’s Claude Code session has it injected before the first prompt. Any agent, any session — same answer.
Works with every agent on your machine
Section titled “Works with every agent on your machine”Auto-wired on sigil init (no manual config):
Claude Code
Cursor
Kiro
Codex CLI
One-line config via sigil register --print:
Continue
Cline
Windsurf
Any MCP client
How memory gets captured
Section titled “How memory gets captured”You don’t call sigil remember manually (though you can). Four hooks run automatically in Claude Code:
| Hook | What it does |
|---|---|
UserPromptSubmit | Searches your memory and injects relevant facts before Claude sees your prompt |
PostToolUse | Captures observations from file edits and shell commands |
Stop | Classifies your last message (preference / decision / constraint / claim) and saves anything memorable |
SessionEnd | Summarizes the session; promotes sticky facts to higher importance |
For other agents (Cursor, Kiro, Codex CLI), memory flows through the MCP server. The agent calls search before answering and ingest when you share something worth keeping.
Honest benchmarks
Section titled “Honest benchmarks”LongMemEval oracle split, n=100, OpenAI top-quality stack:
| Metric | Result |
|---|---|
| R@1 / R@3 / R@10 | 100% / 100% / 100% |
| Search latency (avg / p95) | 33ms / 61ms |
| Cost per full run | $0.21 |
Caveats: oracle is the easy split; n=100 is small; per-question haystack is ~25 chunks. Full methodology and reproducible scripts in eval/longmemeval/.
Get started
Section titled “Get started”docker run -d --name sigil-pg -p 5432:5432 \ -e POSTGRES_PASSWORD=sigil_dev pgvector/pgvector:pg15
npm install -g @anmol-srv/sigilsigil initsigil init asks you two questions (LLM provider, Postgres connection), runs migrations, and wires itself into every AI agent it finds on your machine. That’s it.