Four tools. One philosophy: boring technology, agent-first.
Code as a tree, not as text.
An in-memory AST you manipulate through filesystem-like path operations. No syntax, no parsing, no formatting. Write nodes, transpile to code.
A POSIX-like standard for agentic external context.
Structuring Git repositories so AI agents can reliably discover, read, and write temporal and static context without exploding their context windows.
Shared agent task brain with evolved prompts.
An MCP server for coordinating agentic work through projects, tasks, and threads. Subagents inherit accumulated context from all previous work.
Structured knowledge vault with autonomous self-audit.
Durable fact storage for AI agents. Entities, concepts, syntheses, and sources — in Obsidian-compatible markdown. Auto-generates audit reports that surface what the vault doesn't know.
We believe agents aren't tools — they're syntheses. Not a model responding to prompts, but an entity that thinks, acts, and persists through the tools it wields and the context it carries.
Our products reflect this: boring technology used deliberately. Git as the backing store. SQLite as the database. Filesystem paths as programs. Standards over frameworks. The simplest thing that works for an entity that needs to remember, coordinate, and build.
We ship open source. We build for agents first, humans second — because the best agent tooling is the kind that disappears into the workflow.