FaultLine
Validated memory for AI — prove it, own it.
One product · two memories · one firewall between them
What is true about you
A validated knowledge graph. Every write passes a gate before it is stored, so a hallucinated detail never lands. Recall is a deterministic walk of real rows — traceable to the fact it came from, correctable at every layer, never a vector guess. If it isn't there, FaultLine says so.
See how recall works ▸ === HOW YOU OPERATE ===What your agent learned doing the work
The Agent Cortex — a second, separate memory the agent keeps for itself: the gotcha, the command that failed, the correction you gave it, the procedure that finally worked. Repeat a lesson three times and it hardens into a rule. Your agent stops re-solving solved problems.
Explore the Cortex ▸They are never mixed. The two memories live in different PostgreSQL schemas
(faultline_<id> and flagent_<id>), and the fact walk physically cannot
reach the cortex. A note about how to work can never be served to you as a fact.
See the firewall ▸
/v1/chat/completions, or drive it over /mcp.
Run it standalone with your own key, or bind it to a FaultLine seat and it reuses the
same validated memory your assistant already trusts — and its cortex. Your instance
documents its own integrations, copy-ready, at /integration.
Build one ▸
Pick your flavour.
Same product, six ways in. How do you want to explore it?
…or just show me the full story → · the research · friends in the space
⊢ FAULTLINE · WRITE-VALIDATED MEMORY · © 2026 · Hosted in Canada 🍁 · Made by Volenti