Signet-AI: Local-First Identity, Memory, and Secrets for AI Agents

Signet-AI is an open-source project designed to serve as a portable context layer for AI agents. It manages identity, memory, provenance, secrets, skills, and working knowledge independently of specific chat applications, model providers, or agent harnesses. This architecture allows the underlying execution environment to change while the agent's core context remains consistent.

The project emphasizes local-first storage, providing inspectable provenance and tools for repair. It is compatible with various AI agent harnesses, including Forge, Claude Code, OpenCode, OpenClaw, Codex, Gemini CLI, Pi, Oh My Pi, and Hermes Agent. Signet-AI captures and indexes useful context between sessions, injecting relevant information into subsequent prompts to ensure agents have continuity without manual rebuilding of context.

Signet-AI is built to integrate with ordinary context sources like project notes, transcripts, markdown files, PDFs, and URLs. It offers a semantic layer to help agents navigate this durable record, allowing them to trace back to original sources when summaries are outdated or critical decisions are needed. The project recently released version v0.136.0, which includes a new feature to extract Discord attachment text.