SignetAI | Local-First Memory & Secrets for AI Agents

Memory + secrets, outside the harness

The memory layer
for every agent you use.

One local memory shared by Claude Code, Codex, OpenCode, and the rest. Inspectable, source‑backed, and yours to keep when your tools change.

Install for your operating system
$curl -fsSL https://signetai.sh/install.sh | bash
Get started Join Discord

Portable context across the ecosystem

01Interface

A live console
for your local context.

Every memory, source, and claim is inspectable in the dashboard: graph traversal, recall, and repair without a black box.

02How it works

Remember like a person.
Cite like a librarian.

Signet sits under your agent. It keeps what happened, learns from it while you're away, and brings it back next time.

Three stacked layers read top to bottom: evidence arrives on top, indexes are built from it in the middle, and knowledge sits at the bottom, with threads from each claim up to the evidence it cites.websitedeploycache: sqlitesessions incontext outEvidencekept as receivedIndexesderived · rebuildableKnowledgeentities · claims

03Local by default

Yours to run.
Yours to move.

One binary, a daemon on your machine, and one workspace folder, ~/.agents/. Every drawer has one job, so you always know what to back up and what Signet can rebuild.

  1. The files that say who your agent is and how it works. Plain Markdown and YAML you can open in any editor.

    • AGENTS.md SOUL.md IDENTITY.md
    • USER.md MEMORY.md agent.yaml
    • skills/its own Git repository

  2. Every session lands as a transcript before anything is interpreted. Files you drop in are imported from the inbox.

    • transcripts/<harness>/transcript.jsonlcanonical JSONL
    • files/import inbox

  3. A single SQLite database holds the durable state. Imports keep their original bytes; snapshots are verified before they count.

    • data/signet.dbSQLite, the authority
    • data/imports/retained originals
    • data/snapshots/verified snapshots

  4. Credentials are encrypted on disk. The master key lives in your operating system's keyring, never in the folder.

    • .secrets/encrypted

  5. Logs, locks, and caches are kept apart from everything durable, so Signet can recreate them whenever it needs to.

    • runtime/logs, locks, temporary files
    • cache/rebuildable
The Signet workspace drawn as a cabinet with five drawers: authored files, evidence, data, secrets, and runtime state with cache.05 RUNTIME/CACHE04 SECRETS03 DATA02 EVIDENCE01 AUTHORED
Click a drawer to open it

04Sources

Bring what you
already know.

Connect your notes, docs, repos, and chats where they already live. Signet reads them, never edits them, and every answer cites the exact page, path, or line it came from.

Read about Sources

05Portable context

One context.
Every harness.

Signet installs underneath the tools you already use. Switch shells without starting over; the same memory and secrets follow.

The harness is where the work happens. Memory is what travels.Connector docs

Muse Code (hooks + MCP), opens in a new tabClaude Code (session hooks), opens in a new tabKimi CLI (hooks + MCP), opens in a new tabOpenCode (runtime plugin), opens in a new tabForgeCode (MCP + skills), opens in a new tabSIGNETMEMORYOpenClaw (runtime adapter), opens in a new tabOh My Pi (runtime extension), opens in a new tabCodex (hooks + MCP), opens in a new tabPi (connector), opens in a new tabHermes Agent (memory provider), opens in a new tabGemini CLI (MCP + GEMINI.md), opens in a new tab

06Benchmark

One tracked run.
Receipts attached.

Rules profile, bounded daemon recall, and no model at search time.

97.6%

LongMemEval answer accuracy

MemoryBench rules profile · tracked local and canary score set.

07From the blog

Notes from
building Signet.

Announcements, engineering notes, and essays on how agents should remember.

All posts
Self-Hosted AI Memory for Hermes Agent and OpenClaw

Guide

The Source and the Synthesis

Essay

The Obsidian of AI Memory Systems

Essay

08Get started

Persistent memory
in five minutes.

No cloud account, no config files to write. Install, follow the guided setup, and get back to work.

Install for your operating system
$curl -fsSL https://signetai.sh/install.sh | bash
  1. Install

    One command installs Signet and starts its daemon on your machine.

    signet daemon · running
  2. Set up

    Guided onboarding picks a provider and connects your sources and agents.

    dashboard · localhost:3850
  3. Work as usual

    Your next session starts with context, in whichever harness you open.

    <signet-memory-context>
Read the quickstart Star on GitHub Join Discord

Apache 2.0 · Open source · Runs on macOS, Linux, and Windows