Wardwell began as a Rust CLI I built for my own work. It keeps project state, decisions, and history in files you own, then gives an agent enough context to continue instead of reconstructing everything from chat.
brew tap yakschuss/wardwell && brew install wardwell
A new agent session knows the repository but not the work.
It cannot see why you rejected the obvious approach, what changed yesterday, or what still needs your judgment.
You become the memory layer, repeating your own context to your own tools.
Current state lives in Markdown. History and lessons live in append-only JSONL. The source remains readable without Wardwell.
The MCP server gives agents scoped search, project state, decisions, history, and explicit write paths instead of another pile of chat transcripts.
A session-start hook restores the relevant project context before the first prompt. Work resumes from the last durable state.
Your vault stays on your machine. Agents receive only the domain and project context you expose.
Open the files with any editor, back them up however you want, or stop using the software without losing the record.
Decisions and corrections are recorded explicitly. Generated summaries are useful views, not a replacement for provenance.
If you do not want to configure an MCP server, hooks, or agent workflows, Wardwell.app productizes the same operating model.
The CLI is for builders who want custody and control. The managed product is for people who want the outcome without becoming agent-system operators.