Built like enterprise software. Runs on your own machine.
Two knowledge layers, long-term memory, five-dimension evaluation, and versioned Skills & SOPs — the core that keeps your work grounded and traceable.
Two layers, one brain.
Personal layer
Your private workspace — documents, drafts, past work. Invisible to everyone else.
Company layer
Shared policies, templates and SOPs — one fact base, one set of rules, shared across your team's agents.
Nothing moves up from your layer on its own. Sharing goes through evaluation → authorization → approval — every time.
Memory that compounds.
Every question, correction and decision is remembered as long-term memory. Your assistant doesn't reset between sessions — the longer you work with it, the less you have to explain.
Scored on five dimensions — before it counts.
Score it yourself — or hand it to an independent third-party model.
LLM-as-judge: one AI grades another, and you see the reasoning.
Optional, per request, and you authorise each call.
Evaluation is the only step that can leave your machine — and your knowledge base never does.
Only what passes becomes a skill.
Cleared the bar? It's promoted into your Skills and SOPs — versioned and traceable, so your library stays clean and every result can be traced to its source.
Then your agent runs the work — node by node, inside the process you approved.
Open by design — no lock-in.
Works with OpenAI, Claude & Gemini for optional evaluation.
Swap models any time — nothing your team has learned is locked to a vendor.
A lightweight local model ships in the box, so the basics run with zero setup.
- Files stay on your machine and LAN
- Evaluation is opt-in, per call
- Uninstall means gone
- DPDP-friendly by design
See the core in action.
One month free · No credit card · Files stay on your machine.