Context for the present. Memory for the past. Evidence for what happened.
Open-source, local-first infrastructure for AI agents.

Website · Perseus · Vault · Ledger

An agent can produce an answer and still leave important questions unanswered: what state did it see, what did it remember, and what record shows what happened?

Perseus Computing builds the local-first infrastructure around the model. The products run independently or together, use documented interfaces, and do not require a hosted control plane or a particular agent framework.

In one sentence:

• Perseus answers: What does the system need to know right now?
• Vault answers: What did it learn or decide before?
• Ledger answers: What happened, and can we verify it?

Current integrations should use the Ledger identifiers documented in the repository.

Install the context engine in a project and render its first workspace briefing:

pip install perseus-ctx
cd your-project
perseus quickstart

Then add the layers you need:

• Perseus quickstart
• Perseus Vault
• Vault MCP API reference
• Perseus Ledger
• Integrations

Your project context, memory, and evidence trail should stay under your control. The core products run locally or self-hosted, use inspectable formats, and keep external runtimes optional rather than making them dependencies.

• No required hosted control plane
• No required cloud account for the local path
• Works with MCP-compatible assistants and agent frameworks
• MIT-licensed core projects

The products are built for deployments where boundary control and traceability matter. Actual security and compliance properties depend on how a deployment is configured and operated.

Government and constrained-environment information →

• Perseus methods and benchmarks
• Perseus integrations
• Perseus Computing website
• GitHub Sponsors

perseus.observer · GitHub organization · Questions and discussions

_{Open-source infrastructure for agents that need context, continuity, and proof.}