CAIN-42 CAIN StudioAgent governance · enforcement · evidence
Prove your AI agents were allowed to act.
We put a default-deny enforcement boundary between your agent and its tools. A call that was not authorized does not run. Every decision is signed, chained, and verifiable offline by your buyer — without trusting us.
What it is
A fixed-price review and an enforcement boundary: which tools an agent may call, what stops a bad call, and a signed record of what happened.
What it is not
Not a prompt filter, not a compliance certificate, not a promise that your agent is safe. It is evidence about authorization and execution.
What you get
A report, a default-deny policy, and an evidence bundle your security reviewer or customer verifies with a script that imports none of our code.
Three parts, one execution path
CAIN Studio is the hosted governance platform. MCPGate is the enforcement point inside your environment. CLAWX is the identity, approval and operator interface. You do not need all three to start.
CAIN Studio
HOSTED GOVERNANCE
Evaluates an action against identity, authority, policy, risk and trust, and returns a decision you can verify.
MCPGate
CUSTOMER-CONTROLLED ENFORCEMENT
Runs in your network. Default deny, per-tool policy, single-use capability tokens, signed decision ledger.
CLAWX
IDENTITY · APPROVAL · OPERATOR
Who the agent is, who approved the exact action, and where a person sees what is happening.
The invariant is simple: no authorization, no execution. A tool call that is not explicitly allowed is refused before it reaches the tool.
How an engagement runs
- Send the surface. Tool definitions, policy, or a staging endpoint. Nothing else is needed for the $500 assessment.
- We produce the artifact. Findings, a default-deny policy, and a signed bundle. A person reviews every finding before it ships.
- You verify it. Run the bundled verifier on a clean machine. A tampered copy must fail; if it does not, we have a bug, not a proof.
- You decide. Fix it yourself, have us remediate, or deploy the gate and keep the evidence.
What "verified" means here
The bundle's signatures and hash chain are intact, the report and findings are bound to the signed record, and the tool that produced it exited zero after checking itself. It does not mean a third party audited your system, or that your agent is secure.
Fixed-price offers
Prices are per engagement, in USD. Scope is written before payment. No subscription is required to start.
Focused AI Agent / MCP Security Assessment
One agent or MCP server, up to 20 tools. Tool-surface findings, a default-deny policy, and a signed evidence bundle. 3 business days.
Agent Security & Governance Review
Up to 5 agents and 50 tools. Answers the security questionnaire blocking your enterprise deal, with evidence or an explicit gap. 5 business days.
MCPGate Assessment & Remediation
Deploy the enforcement boundary in your staging environment, build the policy, run the 19-case adversarial harness, hand over the runbook. 10 business days.
Enterprise Agent Governance Engagement
Discovery, architecture review, a pilot in one team, and a costed 90-day plan. Requires a scoping call before any quote is accepted.
Don't trust this page. Verify the artifacts.
Every deliverable ships with a verifier that imports no CAIN code. The public evidence in this repository is checkable now; the customer deliverables are checkable when you receive one.
Run it yourself
python3 verify.py ./bundle --pinned-pub <sha256>
Exit 0 = INTACT. Exit 1 = FAILED or EMPTY. The verifier prints every problem it found.
What is public today
- Verification and limitations — what is proven, what is not.
- The MCPGate package is in the repository under
deploy/mcpgate-standalone. - No third-party audit exists. The claims registry records zero independently verified claims.
Start with one artifact
The $500 assessment is the fastest honest path to a decision. If it is not useful, you have lost three days and $500, not a quarter and a platform migration.
Stated plainly
FIRME AI CORPORATION is a small team. We have no SOC 2, no ISO 27001, no third-party security audit and no enterprise reference customer. No revenue or customer is claimed on this page. The evidence we sell is evidence about authorization and execution, produced by our tools and verifiable by you — not a certification and not a guarantee.