Start free. Pay per agent on ours. Unmetered on yours.
One agent on our hosted console is free. When you want more agents, or to evaluate on every change, the Team tier is priced per agent on the same console. When your agents’ traffic has to stay inside your own network, the platform runs there — VPC, on-premises, or air-gapped — as an annual subscription with no agent count and no token bill.
Free to start, per agent on ours, unmetered on yours
On our infrastructure, one agent is free and discovery is free at any scale; beyond that you pay per agent, never per evaluation or per token. On yours nothing is metered at all: free on the open-source engine, an annual subscription for the platform.
- Runs on
- Your own process
- Bounded by
- Your hardware
- Discover
- The endpoint, on your machines
- Diamond
- —
- Dome
- The engine, in your process
- Darwin
- —
Vijil Dome, the control engine, under Apache-2.0 — read it, fork it, change it. pip install vijil-dome
- Runs on
- console.vijil.ai, at our cost
- Bounded by
- One agent; a baseline evaluation a week, a Quick adaptive run a month
- Discover
- Any scale
- Diamond
- One agent, scored
- Dome
- One agent, enforced
- Darwin
- In development
Not a crippled version of the paid one — found, scored and enforced.
- Runs on
- console.vijil.ai
- Bounded by
- The agents you choose; monthly or annual
- Discover
- Any scale
- Diamond
- Each agent, on every change, with every adaptive effort
- Dome
- Each agent, enforced
- Darwin
- In development
Where a team goes when one agent is not enough. Never per evaluation, never per token.
- Runs on
- Your VPC, datacenter, or a network with no egress
- Bounded by
- Your hardware
- Discover
- Any scale, in your network
- Diamond
- Your policy, on a cadence
- Dome
- Every agent, enforced
- Darwin
- In development
Every agent, every evaluation. No count, no token meter.
After the open-source engine, each column is the next step: one agent free, then the agents that matter on Team, then the whole population in your own network once its traffic cannot leave.
Four products, three places to run them
The first column is what runs on your own machine for nothing. Only Dome and the Discover endpoint are open source — the Diamond client is free to install but is not, and that is worth saying here rather than leaving to a license file.
| On your machine | Hosted — free | Hosted — Team | In your network | |
|---|---|---|---|---|
| Runs on | Your own process | console.vijil.ai, at our cost | console.vijil.ai | Your VPC, datacenter, or a network with no egress |
| Bounded by | Your hardware | One agent; a baseline evaluation a week, a Quick adaptive run a month | The agents you choose | Your hardware |
| Discover | Installers for macOS, Windows and Linux, plus a browser extension. Open source. | Every GitHub org and cloud account, scanned as often as you like | Every GitHub org and cloud account, scanned as often as you like | Every surface, including what a hosted scan cannot reach |
| Diamond | pip install vijil-sdk — evaluate, harnesses, reports. Free, not open source. | One agent scored against the Vijil baseline harness | Each agent, on every change, with Quick, Standard and Deep adaptive runs | Every agent, scored as often as you like. No per-evaluation meter |
| Dome | pip install vijil-dome — guards on inputs, LLM calls, tool calls, outputs. Apache-2.0. | One agent enforced, with what its guards blocked and their latency in the console | Each agent enforced, with what its guards blocked and their latency in the console | Every agent, with decision traces kept in your network. No per-agent count, no token meter |
| Darwin | In development, with design partners. Become one | — | — | — |
Answered, including the ones that do not flatter us
So what does it actually cost?
For one agent on our console, nothing. On Team, a price per agent, monthly or annual, with no per-evaluation meter and no token bill. In your own network you buy a deployment rather than a quantity: an annual subscription for the platform, with no agent count. Services publishes the shape of the engagement that gets you there — what it does not publish is a number, because the scope of a first deployment is the thing we work out together.
Is the free tier self-serve?
Not yet. Today you ask and we open an account, usually the same day. The open-source engine needs nobody’s permission and you can have it running in the next ten minutes; the hosted platform still goes through us.
What exactly is open source?
Vijil Dome — the control engine — under Apache-2.0. That is the whole answer today. The CLI and the Python client install freely from PyPI, and their source is not yet published.
Three more, answered the same way
Do we have to send you our data?
On the hosted tiers, free and Team, yes — they run on our console, so evaluation traffic and decision traces reach us. In your own network, no: the platform runs inside your VPC or on your own hardware, and it is designed to run in a network with no egress at all. That is the reason that tier is unmetered: we cannot count what we cannot see. We do not train on customer data — not on the free tier and not on any other. What we will do is help you train on your own data inside your own network, where it never egresses. Our Trust Center carries the security program, and the privacy policy names every subprocessor and where each one processes.
What does the free tier not do?
It governs one agent on our console, so it cannot show you the failure that only appears between two of them — which is the failure this company exists to address. What it can finish: a baseline score with the transcripts behind every failure, and one agent enforced end to end. Those are whole answers, not samples. It runs one baseline evaluation a week and one Quick adaptive run a month; evaluating on every change, more agents, and the Standard and Deep efforts are Team. Until signup is self-serve, we open each free account and hold those limits by hand. It holds recent history rather than a year of it, and it will not export. Nothing about it expires.
Why is discovery free at any scale?
Because a count you cannot obtain is worth nothing to you, and an agent we cannot see is worth nothing to us. Metering visibility would mean charging you to find out how much you owe — and in your own network there is nothing to owe, because the inventory never leaves your network.
Score one agent you already trust
Not the one you expect to fail — the one you were sure about. If it tells you something you did not know, the conversation about the rest is a short one.