Pricing
Start with evidence about your own agent
A free snapshot of five write tools, then a fixed-price review of one agent’s production write paths. A continuous gate is available by invitation after a review.
Sandbox
Simulation only
Free
No agreement required
- 1,000 evaluations / month
- Mock responses — not production
- Public API + Governance Console
- Community support
Pilot snapshot
Free · limited founding availability
Free
Scope and availability confirmed on a call
- One agent, five named write tools
- Which actions could run unattended, which need approval
- A short written summary
- Founder-led, from read-only material
Write-access review
Fixed price · one agent
$4,500
50% deposit after agreed scope; delivery date reserved on a call
- Up to 10 named writes, one environment/provider
- Up to 200 supplied records in one format
- Whether each can be undone, from evidence
- Past writes risk-scored, where they map to ARF's action model
- Unattended vs. approval-required, action by action
- Explicit gaps and next steps
What each step includes
The difference is depth: the snapshot looks at five tools, the review maps every write path in the agreed tool inventory. Neither needs write access to your systems.
| Capability | Sandbox | Pilot snapshot | Write-access review |
|---|---|---|---|
| What you get | Simulated decisions | A five-tool write-access map | A map of up to 10 agreed write operations |
| What we need | Synthetic or redacted payloads | Tool list + representative traces | Tool definitions + action history |
| Access to your systems | None | Read-only material only | Read-only material only |
| Enforcement | Simulated only | None — analysis only | None in the review; a gate is by invitation afterwards |
| Deliverable | API responses | Short written summary | Report, recoverability map, scored history, gap list |
| Support | Community | Founder-led | Founder-led |
| Commercials | Free | Free, scope and feedback agreed first | $4,500 fixed; 50% deposit; card or invoice |
How it works
ARF is not sold per seat. You start with evidence about your own agent, and pay only for the depth you need.
01
Start with a snapshot
Five named write tools, analysed from read-only material within an agreed short scope. Limited founding availability, confirmed on a call. Honest feedback is welcome; a public reference is optional.
02
Then a fixed-price review
One agent's production write paths, mapped action by action: which can run unattended, which should wait for a human, and what evidence remains when it acts.
03
A continuous gate, by invitation
After a review, ARF can gate your agent's write path where that path is modeled — demonstrated today for ONTAP storage operations against a simulated cluster. Scoped and priced per engagement.
Common questions
What does the review deliver?
A review of one agent, up to 10 named write operations in one environment/provider, and up to 200 supplied records in one format. You receive an inventory and recoverability map, a decision memo, supported findings and evidence index, one readout and a factual correction pass. Scoring is conditional on a supported action model; unmapped writes and unknowns are listed. The review does not certify the agent or install a production gate.
Do you need write access to our systems?
Not for a snapshot or a review. Both use read-only material: tool definitions, representative traces or action history, and any existing policy. A continuous gate, if you choose one later, does sit in your agent's write path, and is scoped separately.
Which agents is this for?
Teams whose agents are moving from read-only recommendations to production changes — infrastructure and storage operations first. Agents that make other consequential changes, such as customer-facing commitments, can be scoped on a call.
What happens after the review?
Nothing automatic. No subscription starts and nothing else is billed. If you want the decisions enforced rather than recommended, a continuous gate is scoped separately.
Is the sandbox safe to point at production data?
No. The sandbox returns simulated responses and should be treated as a demonstration surface. Use synthetic or redacted payloads.
How do we pay?
Card or a simple invoice. The review is $4,500: $2,250 before reserving work after scope is agreed, and $2,250 due seven calendar days after delivery under the agreed proposal. Start and delivery depend on complete inputs and reserved capacity. Cancellation, handling and acceptance terms are agreed before payment; additional tools, formats or providers need a new scope.
Start in the sandbox. Bring your agent when you are ready.
Tell us which agent, the production writes it makes or wants to make, and who is asking for controls. We reply within two business days.