Agent Assurance Engine

Pricing

Build robust agents with Assurance Engine. Start free and scale with your fleet.

10 agents

Free

Find how your agent can break its own rules, and fix it before you ship.

$0 / mo
Start for free
  • 3 agents
  • Scans your prompts, tools, and code for the ways your agent can go wrong
  • Turns each risk into a repeatable test, up to 7 per agent
  • Fix and re-test from Claude Code, Cursor, or Codex
  • $25/mo in credits for scans and tests
  • 90 days of history

Team

Keep testing in production, and switch models with evidence.

$399 / mo
Request Team onboarding
  • 10 agents, 5 more for $149/mo
  • Everything in Free, with up to 30 tests per agent
  • Runs your tests against the live agent as often as hourly
  • Shows which rule broke, how often, and since when
  • Compares models on the same tests before you switch
  • Tunes agent settings by experiment and opens the winner as a draft pull request
  • $440/mo in credits

Enterprise

Run Assurance Engine inside your own network.

Custom billed annually
Contact us
  • Unlimited agents
  • Everything in Team, with custom test frequency
  • Runs in VAIL cloud, your cloud, or on-premises
  • Can use a local model, so your source never leaves your network
  • A named engineer and an SLA

Unlimited team members on every plan. Scheduled tests never use credits.

Compare plans

Feature Free Team Enterprise
Coverage
Agents310, then 5 more for $149/moUnlimited
Team membersUnlimitedUnlimitedUnlimited
Tests per agentUp to 7Up to 30Up to 30
Agent frameworksAll public platformsAll public platformsAll public platforms
Connect source from GitHub or an uploadYesYesYes
Find and fix
Scan for the ways your agent can break its rulesYesYesYes
Risks mapped to OWASP, MITRE ATLAS, and NISTYesYesYes
Marks which risks depend on the modelYesYesYes
Fix and re-test from your coding agentYesYesYes
Compare models on the same testsNot includedYesYes
Tuning experiments that open a draft pull requestNot includedYesYes
In production
Scheduled tests against the live agentNot includedHourly, daily, or weeklyCustom frequency
Findings: which rule broke, how often, since whenNot includedYesYes
Send results to LangSmithNot includedYesYes
Reach agents inside a private networkYesYesYes
Plan
Monthly credits for scans and tests$25$440Committed
History90 days90 days90 days
Where it runsVAIL cloudVAIL cloudVAIL cloud, your cloud, or on-premises
SupportCommunityEmailNamed engineer and SLA

Credits and billing

Scheduled tests are included. Tests that run on a schedule against your plan's agents never use credits.

Credits cover work you start yourself: scans, generating tests, re-testing a fix, and tuning experiments. They refresh each month, and Team carries over up to $200 unused.

Nothing upgrades on its own. When credits run out, new scans and test runs pause. Scheduled tests keep running for 72 hours, then pause too.

FAQ

What is an assured agent?

An agent whose own rules have been turned into tests. You know how it can fail, you have fixed what did, and you find out when it starts failing again.

Question A typical agent An assured agent
How it can fail Found when a user runs into it Listed before you ship, each one citing the instruction it would break
Tests Evals for the cases someone thought to write Generated from the agent's own instructions and tools, scored by fixed checks instead of an LLM judge
Fixes A prompt edit and a few manual tries The same tests run before and after, so you can see whether the fix held
Changing models Swap it and watch production Run the same tests on the new model first
In production Traces to read after something goes wrong The same tests on a schedule, with a record of which rule broke and since when

Across 15 agents we scanned, 144 of 153 risks depended on the model's judgment rather than on code, which is the part unit tests do not reach. Read the case study

What counts as one agent?

One agent you have registered, with its own source, tests, and history. Subagents inside it count as part of the same agent.

How many credits does a scan or a test run use?

It depends on the agent: how much source it has, how many tests it has, and how many times you run them. So there is no fixed price per feature. Credits are only used for work you start yourself, and scheduled tests never use them. The app shows what each agent's runs have cost.

Which agent frameworks work?

Scans and test generation work on all public platforms. Running tests against a live agent works today with LangGraph, Hermes, OpenClaw, and any agent behind an HTTP endpoint. Fixing and re-testing locally works with LangGraph, Hermes, and OpenClaw.

Where does our agent's source go?

It is stored for your organization only, and sent to a model provider during scans and test generation. No SDK or trace access is needed. On Enterprise you can self-host with a local model, so it never leaves your network.

What happens after we sign up?

We email you a sign-in link. You can add agents and upload source right away. We review each new organization before its scans and tests run.

Have questions?

Ask us about plans or your setup.

Get in touch