Write-validated memory · over MCP

One memory. Every system you use.

FaultLine is one validated memory that plugs into Claude, OpenWebUI, or any MCP client. Tell it once in one tool — it knows in all of them. Your memory follows you from tool to tool, and every answer is a real stored fact you can prove.

Start free ▸ See the architecture

Free individual tier · no card · hosted in Canada 🍁 · your data kept separate from everyone else's

many clients · one memory

Claude OpenWebUI Your editor Any MCP client

⊢ FaultLine

one validated graph · per person

It doesn't guess. It knows.

Ask your AI anything about you. It answers from real rows — not a statistical guess.

Every fact is checked on the way in and looked up — not guessed — on the way out. When FaultLine is sure, it says so plainly. When it's less sure, it says that too. It never invents a fact it doesn't have.

See how it works

recall_memory · faultline

recall_memory "what do you know about my family?"
↳ grounded facts, traced to when you said them
You live in Toronto with your partner Diane.
You have a son named Elliot, who is 9.
Your mother is Marla; she's 64 and lives in Ottawa.
↳ held tentatively (inferred, not yet confirmed)
You may have a dentist appointment on the 14th.
✓ every line points to a real stored fact.

Six things FaultLine does

Pick what you came for.

Each one is real, shipping, and verifiable against the code. None is a roadmap promise.

⊢ the point

One memory, every model

Plug FaultLine into Claude, OpenWebUI, or any MCP client. Tell it once in one tool — it knows in all of them. Your memory follows you from model to model.

How MCP works ▸

✎

Corrections

You're the source of truth. Say "no, that's wrong" and FaultLine rewrites the record — the stale fact is archived, not buried. No model ever quietly overrules you.

See corrections ▸

⤢

Expand a subject

Teach it a whole domain. /learn grows the ontology — networking, law, biology, anything — so a subject's full depth lands walkable, not flat.

See /learn ▸

◉

Cortex memory

Your agent gets its own private notebook, walled off from your memory. Save a gotcha or a correction once — by hand, or have the agent save one — and it's there next time, instead of re-learned the hard way every session.

See the cortex ▸

▣

Agent chat

Drop a chat agent with real memory onto any page. Use it standalone, or wire it to your FaultLine memory. A banner always says when a conversation is being saved.

See the chat agent ▸

⌘

Bring your own brain

Connect your own LLM endpoint and model. FaultLine is model-agnostic — swap Claude for any model tomorrow, and keep exactly the same validated memory.

See BYO brain ▸

Start free — all six ▸ or see the prices · log in

New here?