The offering
One small agent. Two ways to run it.
The FaultLine chat agent is a small, embeddable widget you can place on any page. It ships in two modes, and both are real products — not a trial leading to the other. Pick the one that fits, and switch any time.
Standalone — your own brain
Point the agent at any OpenAI-compatible endpoint with your own key. It behaves like any normal chatbot, fully usable with zero FaultLine involvement. Your key stays on the agent; nothing is sent to FaultLine.
This is the whole product on its own. Bring it home, host it yourself, never talk to us.
Through FaultLine — one click
Click through FaultLine and the agent binds to one of your FaultLine seats. It then reuses the same brain as that account and reads/writes its validated memory — the same grounded, traceable facts your assistant already trusts over MCP.
No second brain to configure, no second memory to drift. The hardening — fail-closed validation, the confidence firewall — is inherited from FaultLine.
Clear by design
Two choices. They are kept separate on purpose.
People get burned by "memory" products that quietly conflate two different questions. The agent asks them separately, in plain language, so you always know what you're agreeing to.
Memory side — where memory lives
Where the conversation's memory is kept, if at all.
- FaultLine — validated, structured facts held at FaultLine. The bot remembers what's true about you and your world across sessions.
- other — a memory store you bring. The agent passes chat through; it does not manage this memory itself.
- no memory — stateless chat. Nothing is retained or learned; each turn stands alone, the bot starts fresh every time.
Model brain — what answers
Which model actually produces each reply.
- BYO key — your own key and endpoint (standalone). The key stays on this agent; nothing is sent to FaultLine.
- through FaultLine — your bound seat's brain. Chat flows: agent → FaultLine → the seat's brain → response. The brain key never leaves FaultLine.
No quiet remembering
You always know when the bot is learning.
Most memory tools save what you type silently. The FaultLine agent refuses to. The moment a bound seat is in train mode — meaning the conversation is being written to memory — a hard red banner sits across the top of the chat:
No banner means nothing is retained — not by FaultLine, and not quietly. In serve mode the bot reads the brain but writes nothing. Switching a seat between train and serve is a single control in the FaultLine console, and the banner follows it within seconds.
Keys, traffic, and trust
Your model key goes exactly where you tell it.
The two security stories are short, and they depend only on the mode you chose above — no hidden third option.
| Mode | Where chat goes | Where the key goes | What's retained |
|---|---|---|---|
| BYO key | Agent → your endpoint → reply | Stays on the agent. Never sent to FaultLine. | Depends on the memory side — possibly nothing at all. |
| Through FaultLine | Agent → FaultLine → your seat's brain → reply | Stays at FaultLine. The agent never sees it. | Validated facts in your FaultLine seat — only when the banner is shown. |
| No memory | Whichever brain you chose | As above | Nothing. The bot never retains anything. |
Try it first
A 15-day free trial. No auto-subscribe, ever.
The chat-agent feature has its own trial, separate from any FaultLine trial. Click start free trial from the agent's connect screen and the feature turns on for 15 days.
What happens at the start
The feature turns on and you get a brief intro walkthrough the first time you connect or subscribe. An email confirms the trial and its window.
What happens at the end
The feature turns off. You are not charged, and nothing flips you onto a paid plan on its own. A reminder email lands before it expires.
If you want to keep it
A Subscribe button is available during and after the trial — in the FaultLine console and in the email. Subscribe when you're sure, on your timing.
The add-on
Chat-agent pricing — low, and on top of the seat.
This is the charge for the chat-agent feature — the widget, the through-FaultLine capability, and the control plane. It's separate from, and on top of, the FaultLine seat. Because you can bring your own brain, FaultLine's marginal cost to serve chat is close to zero, so the add-on is priced low. Prices in CAD, billed monthly.
- 1 chat agent
- Serve (chat) included
- Standalone or through FaultLine
~$40 per agent
- 5 chat agents
- Serve (chat) included
- Operator-provisioned bucket
~$32 per agent
- 10 chat agents
- Serve (chat) included
- Operator-provisioned bucket
~$32 per agent
The agent seat underneath (the memory brain) is billed separately — see the full price list. Train (ingest) is metered through the existing train/serve lever, with quota included and credit-pack overage.
Straight answers
FAQ
Does the chat agent need a FaultLine account?
No. It runs standalone with your own model key and endpoint, fully usable with zero FaultLine involvement. If you later click through FaultLine, it binds to a FaultLine seat and reuses that seat's validated memory and its brain. Both paths ship in the same widget.
What are the two choices I keep hearing about?
They are kept separate on purpose. The memory side decides where the conversation's memory lives: FaultLine, a store you bring, or no memory at all (stateless). The model-brain side decides which model answers: your own key, or through FaultLine. You can mix them — no memory + BYO key is a fully stateless bot.
Where does my model key go?
With your own key, the key stays on the agent and is never sent to FaultLine. Through FaultLine, the agent never sees the brain key at all — chat flows agent → FaultLine → your seat's brain → response, and the key stays at FaultLine the whole time.
How do I know when the bot is remembering what I type?
A hard red banner across the top of the chat window reads INGEST IS ON — this conversation is being saved to FaultLine memory whenever a bound seat is in train mode. No banner means nothing about the conversation is being retained.
Is the trial really free, and will I be charged when it ends?
The chat-agent trial is 15 days and there is no auto-subscribe. When it ends the feature simply turns off — you will not be charged. A Subscribe button is available during and after the trial if you decide to keep it.
Can I embed it anywhere?
Yes. The same agent ships three standard ingress paths: an embeddable web widget (a script tag you drop on a page), an OpenAI-compatible /v1/chat/completions endpoint, and an /mcp server. Whichever mode you choose, all three keep working.
Give your chat real memory — or don't. Your call.
Start standalone and stay there forever, or click once to bind the same validated memory your AI already trusts. Fifteen days free, no charge at the end.