Code is increasingly written by agents. We build tools to make
that work better: context infrastructure, autonomy loops, and
orchestration. Open source and self-hosted.
projects
trieactive
Self-hosted context infrastructure for coding agents.
The host is your git repo: a meaning + intent index as plain
Markdown, next to your code. What every symbol does, why it is that
way. Kept fresh by sync, kept honest by a commit gate. No services.
Desktop editor for agentic engineering, built on trie + opencode.
Includes the attention-map research (AGM): a live view of where a
coding agent is thinking, turn by turn.
Agentic browser automation that compiles cached reflexes: an agent
explores a website once, warm runs replay with zero LLM calls, and
humans join the loop programmatically when the web pushes back.
A self-improving swarm of autonomous agents (supervisor, workers,
reviewers, testers) coordinating through message passing to build
software, including their own codebase.
Parallel Claude Code development: multiple agent instances working
features simultaneously in isolated worktrees and tmux windows,
coordinated by a supervisor.
Declarative multi-step LLM workflows in a YAML-like syntax: chain
prompts, fan out in parallel over iterables, reduce back.
Split-map-reduce for inference.