How it works
Not a wrapper. An engine.
Most AI tools just pass your prompt to an API and hope for the best. Soleri is the other 95%: the six systems that make the model's output actually useful.
Six systems, one engine.
Turn on what your agent needs. Leave the rest off.
The Vault stores what your agent learns. The Brain figures out what's worth keeping. The Curator cleans up the mess as knowledge grows. The Planner gives work structure. Memory carries context between sessions, and Governance sets the quality rules. Run all six or just the ones you need.
It remembers what worked.
Next time similar work comes up, the good stuff shows up on its own.
Fixed something last week? When a similar problem shows up, your agent already knows the approach that worked. You don't have to search for it or explain it again.
It stays clean on its own.
Three systems keep things from turning into a junk drawer.
The Brain tracks what actually gets used, so useful stuff rises and stale stuff fades. The Curator catches duplicates and contradictions on its own. And Governance lets you control what gets shared with the team versus what stays private.
Nothing leaves your machine.
Local. Open source. Private by default.
All of it stays on your computer. No cloud, no tracking, nothing phones home. Apache 2.0.
Try it.
One command to start. Works with Claude Code, Cursor, Codex, and OpenCode.