SignetAI | Portable Agent State
PORTABLE AGENT STATE

Identity, memory, and secrets that survive the model switch.

Signet keeps agent state outside any single model or harness. It preserves the exact record, builds useful context from it, and keeps credentials out of the model window.

Signet Agent Overview — dashboard showing agent health, memory scoring, and session insightsSignet Skills Marketplace — browse and install AI skills with trust profiles and reviewsSignet Ontology Explorer — constellation graph view of knowledge entities and relationships
02 / 04
FIG_01 // AMBIENT_FLOW
ACTIVE_STREAM

Ambient Memory

Every session is automatically distilled into structured memory. No tool calls, no manual saves — your agent wakes up knowing what happened yesterday.

FIG_02 // KNOWLEDGE_GRAPH

Knowledge Graph

Entities and relationships extracted with high precision. Graph traversal allows for deep context retrieval.

FIG_03 // INSPECTION_UI

Inspect Everything

Memory is not a black box. Browse, search, and edit through a refined dashboard with full provenance.

FIG_04 // CROSS_HARNESS

Cross-Platform

A unified agent state layer that keeps identity and memory coherent across every major AI agent harness.

Claude Code OpenCode OpenClaw Codex Forge
FIG_05 // CRYPTO_LAYER

Encrypted Secrets

Industry-standard encryption at rest. Agents utilize credentials without ever exposing plaintext.

FIG_06 // MULTI_TENANT

Multi-Agent

Granular visibility controls. Define isolated, shared, or group memory scopes for swarms.

FIG_07 // CONTEXT_RANKING

Context Ranking

Inspectable ranking, provenance, and dampening keep useful context visible without flooding the model window.

FIG_08 // EXTENSIONS

Core Ecosystem

SDK & React Hooks CLI Interface Git Provider Sync Skills Marketplace Docker / Self-Host MCP Support
03 / 04

Early results on the LoCoMo long-conversation benchmark. Only local-first, open-source system in the top 10.

87.5% Answer Accuracy LoCoMo long-conversation benchmark, 8-question full-stack sample
100% Hit@10 Retrieval Every relevant memory found in the top 10 candidates
0.615 MRR Mean Reciprocal Rank — how high the right answer appears
LOCAL No LLM at Search Time Retrieval and ranking run entirely on-device

Larger evaluation runs in progress. These numbers will change.

Methodology and full results →
04 / 04

Your agent has memory in five minutes.

01

Install

$ bun add -g signetai
or npm i -g signetai
02

Setup

Run signet setup, pick your harness, and the daemon starts automatically. Hooks are installed. Identity files are created.

03

Work

Open your coding agent and start a session. Memory extraction happens automatically. Next session, context is already there.