Each one came out of a problem we had. Some are open source. Some we host. Ask us about any of them.
Memory for AI coding agents. Indexes every Claude Code, Gemini CLI, Codex, OpenCode, Antigravity and Cursor session, then lets the agent search and resume its own history through MCP, backed by a temporal knowledge graph.
A structural code-intelligence engine that runs as an MCP server. Tree-sitter across 40+ languages, a trigram index and a dependency graph, so an agent calls a tool instead of reading whole files.
Web fetching and extraction built for agents. Escalates through three layers (HTTP, a lightweight browser, then challenge solving) to get clean markdown out of pages that resist it.
Context distillation for long agent sessions. Cuts what an agent has to carry between turns so conversations stay coherent and affordable instead of growing without bound.
Reads an AWS, GCP, Cloudflare or OVH account through a read-only role and says where the money goes: idle resources, previous-generation instances, commitments worth making. One library, exposed over both MCP and HTTP.
Runs LLM inference on the cheapest GPU it can find. A Kubernetes controller buys spot capacity across seven providers, serves vLLM on it, and gives the machines back when the work stops. A second repository deploys the whole stack onto hardware you already have.
A desktop app that runs coding agents as a team. Work items outlive any one session, each role (architect, coder, reviewer, tester) is pinned to its own agent CLI and model, and playbooks run them in sequence.
We are building more, and we run an open beta on most of them. Ask us about any of our products.