How Signet Works — SignetAI

How Signet Works

Signet gives AI agents persistent memory, portable identity, and encrypted secrets — all running locally on your machine, across every platform you use.

Overview 01 / 06
1
Configure
~/.agents/agent.yaml — plain YAML you can read and edit by hand.
2
Remember
signet remember — auto-embedding into local SQLite, no cloud required.
3
Recall
Semantic + keyword search, ranked by importance and recency.
4
Sync
Daemon regenerates platform configs — CLAUDE.md, .cursorrules, etc.
5
Carry
New machine? signet init --import. Fully restored in seconds.
Data Flow 02 / 06

Memory Pipeline

  1. Harness hook fires (session-start, user-prompt, or session-end)
  2. Connector calls daemon HTTP API
  3. /api/hooks/remember enqueues extraction job
  4. Extraction worker: LLM decomposes raw input into discrete facts and entity triples
  5. Decision worker: evaluates each fact against existing memories (add / skip / bounded repair)
  6. Controlled writes: new memories inserted with dedup and safety gates
  7. Knowledge graph + embeddings: structured memory substrate persisted for later recall
  8. Embeddings computed and stored alongside each memory
  9. /api/memory/recall builds a bounded candidate pool (graph + BM25 + vector)
  10. Predictive scorer reranks when available, and fails open to baseline ordering when it is not

Config Sync

  1. User edits ~/.agents/AGENTS.md (or any identity file)
  2. File watcher detects the change via chokidar
  3. 2-second debounced sync: regenerate CLAUDE.md, .cursorrules, etc.
  4. 5-second debounced git commit: version-control the change automatically
  5. All connected harnesses receive updated config on next session
Memory Pipeline 03 / 06
Stage 01

Extraction

A local LLM decomposes raw conversation content into discrete facts and entity triples. Each fact is a standalone piece of knowledge; each triple captures a relationship between entities. No data leaves your machine.

Stage 02

Decision

Every extracted fact is evaluated against your existing memory store. The decision worker determines the right action — add a new memory, update an existing one, mark one for deletion, or skip entirely. No duplicates, no noise.

Stage 03

Controlled Writes

Safety gates and deduplication logic ensure only meaningful changes reach the database. Writes are atomic — partial failures never corrupt your memory. Each write is versioned for full audit history.

Stage 04

Knowledge Graph

Entities and their relationships are persisted as a graph structure alongside flat memories. This enables graph-augmented search — finding connections between concepts that keyword or vector search alone would miss.

Stage 05

Retention

Automatic cleanup of expired or low-value data keeps the memory store lean. Retention policies are configurable — you control how aggressively stale memories decay and when they get pruned.

Stage 06

Shadow Mode

Observe exactly what the pipeline would do before enabling writes. Shadow mode runs the full extraction and decision pipeline but commits nothing. Useful for tuning, debugging, and building trust in the system.

Platform Connectors 04 / 06
Connector

Claude Code

Generates and maintains CLAUDE.md from your identity files. Session hooks fire on start, prompt, and end — feeding conversations into the memory pipeline automatically.

Connector

OpenCode

Syncs AGENTS.md into your OpenCode configuration. The runtime plugin handles session lifecycle and memory injection without manual setup.

Connector + Adapter

OpenClaw

Patches OpenClaw config for harness integration. The runtime adapter calls the Signet daemon for memory search and secret injection during active sessions.

Coming Soon

More Platforms

Cursor, Windsurf, and other editors are on the roadmap. The connector architecture is designed for new platforms to be added with minimal effort.

Write once, appear everywhere. Connectors are idempotent — safe to run as many times as needed. You define your agent's identity, memory, and skills in ~/.agents/. The daemon handles syncing to every connected platform.
Cryptographic Identity 05 / 06

Signet is more than a memory layer. It is a foundation for agent identity that no single entity controls. Your agent is a reflection of your expertise — its memory, skills, and personality belong to you. Current AI platforms chose vendor lock-in, barring access to your own customizations behind paywalls. Signet chose a different path.

Cryptographic Identity

A portable trust layer for verifying agent identity and authorization across environments. The goal is strong verification without collapsing back into a central platform.

Agent-to-Agent Discovery

Agents should be able to discover and interact with other agents through open protocols. Trust is established cryptographically, not through platform gatekeepers.

Encrypted Sync

Access your agent across environments with encrypted state and user-controlled keys. Identity, skills, secrets, and memories should stay yours to unlock.

Open Agent Economies

When agents can discover and trust each other, they can transact. Agents hire other agents for real services. Standardized identity makes this possible across every industry.

Decentralized Ownership

No single company or entity should restrict access to your own agent. An LLM is just an LLM — what makes it yours is the knowledge, skills, and identity you build around it.

Corrigibility as a Living Standard

The trust layer is built in. Signet tracks mistakes and works to prevent them from recurring. Observability and accountability are first-class concerns, not afterthoughts.

Your agent is yours.
Dive Deeper 06 / 06

Architecture

Full technical architecture — packages, data flow, and system design.

Read more →

Memory Pipeline

Deep dive into extraction, decision, graph, and retention stages.

Read more →

Platform Integrations

How connectors work across Claude Code, OpenCode, OpenClaw, and more.

Read more →

All Documentation

Setup guides, API reference, CLI commands, and everything else.

Read more →