Pricing
Priced on what Arc protects — not how many seats you have.
Arc is the spend-and-action guardrail for AI agents. Every action passes an allow / ask / block policy, waits for a human when it’s risky, runs only as a signed request your app verifies, and lands in a redacted, hash-chained audit log — with a cumulative spend cap so no agent can quietly run up the bill.
Start free — sign up for a hosted workspace, no credit card. We meter the two things that scale your risk: protected agents and guarded actions — never per seat.
- Free Developer tier — no card
- Metered on risk, not seats
- Month-to-month, no lock-in
Plans
Four tiers. One product.
Every tier ships the whole trust envelope — policy, approval, signed execution, audit. Higher tiers raise the meters and the guarded-spend pool, and unlock retention, export, and streaming.
Developer
Free hosted
Free no credit card
- Protected agents
- 1
- Guarded actions / mo
- 10,000
- Guarded-spend pool
- $10k
- Audit retention
- 7 days
Protect your first agent. A hosted workspace and the @geostack/arc SDK — zero cost.
Team
Most teams start here
$99 per month
- Protected agents
- Up to 10
- Guarded actions / mo
- 250,000
- Guarded-spend pool
- $250k
- Audit retention
- 30 days
Agents in production touching refunds, sends, deletes, or spend.
Business
Compliance pressure
$499 per month
- Protected agents
- Up to 50
- Guarded actions / mo
- 2,000,000
- Guarded-spend pool
- $2M
- Audit retention
- 90 days
Several agents across products, with an audit trail to prove it.
Enterprise
Large / regulated
Custom by quote
- Protected agents
- Custom
- Guarded actions / mo
- Custom
- Guarded-spend pool
- Custom
- Audit retention
- Extended
Single-tenant deployment and an SLA today; SSO, SIEM streaming, and a third-party security review on the roadmap.
Agent and action limits are soft ceilings, not hard cut-offs — Arc never silently blocks your production traffic because you crossed a line on a pricing page. If you’re consistently over, we’ll move you to the right tier. See What happens if I go over? below.
Every plan includes
The whole trust envelope. It’s the product, not the upsell.
Free or Enterprise, every Arc plan ships the full control loop. Paid tiers raise the meters and add retention, export, and streaming — not the guardrails themselves.
- Allow / ask / block policy
- Human approval on ask
- Signed (ES256) execution
- Redacted, hash-chained audit
- Cumulative spend / budget caps
- TypeScript SDK + CLI
- MCP adapter
- Dashboard + API
How metering works
Two meters, both tied to risk — never seats.
Seat-based pricing punishes you for adding teammates and tells you nothing about exposure. Arc meters the two things that actually track how much risk you’ve handed to agents.
A registered agent that made at least one governed call during the month. It’s the “active identity” meter — your seat-equivalent, except it counts machines that can act, not humans who can log in. An agent that sat idle all month doesn’t count.
One policy decision — a single allow / ask / block evaluation on an action an agent tried to take. A refund, a delete, an outbound email, a prod write: each is one guarded action. Blocked and asked actions count too — stopping the bad ones is the work you’re paying for.
$ guarded-spend pool
not a charge · not a wallet
It’s the headline of how much agent spend Arc stands guard over each month under your plan — “Arc is guarding up to $250k/mo of agent spend for you.” We deliberately don’t price the spend cap itself, because budget caps are now commodity inside free gateways (LiteLLM, Portkey, Bifrost). You’re paying for the trust envelope gateways don’t have — pre-action human approval, signed execution your app verifies, and the hash-chained audit trail — and the spend cap rides along inside it.
What you’re actually paying for: a free gateway can cap a budget; it can’t prove who approved the $40k refund or hold a signed, append-only record that it happened.
Why this exists
The cap that doesn’t exist is the one that costs you.
Reportedly ~$500M on Claude in one month, because no one set a cap.
An AI consultant told Axios that one enterprise client reportedly spent around $500M on Claude in a single month — because they set no usage caps and token spend exploded across thousands of people running agentic workflows. No company has confirmed the figure, so treat the number as a story, not a stat. The pattern around it is on the record: Microsoft cancelled most internal Claude Code licenses, and Uber said it exhausted its 2026 AI budget by April.
The common thread isn’t a rogue model. It’s that the cap either didn’t exist or was never turned on. Arc is that cap — plus the part gateways skip: the approval, the signed execution, and the audit trail that prove a destructive or expensive action was authorized before it happened.
Arc guards spend and the irreversible actions a budget line never sees — the refund, the delete, the prod migration, the mass email.
Compare
Where the tiers differ.
The guardrails are identical on every plan. What changes is operations & compliance — retention, export, access control, and how audit data leaves Arc.
| Operate & comply | Developer Free | Team $99/mo | Business $499/mo | Enterprise Custom |
|---|---|---|---|---|
| Audit export | JSON / CSV | Signed / verifiable | Signed / verifiable | |
| Approval channels | In-app | In-app · email · Slack | + webhooks | + custom routing |
| Org / workspace controls | Included | Included | Included | |
| Role-based access | Basic | Included | SSO seats (roadmap) | |
| Key rotation | Included | KMS options (roadmap) | ||
| SIEM streaming | Add-on (roadmap) | Roadmap | ||
| Deployment | Hosted | Hosted | Hosted | Single-tenant |
| Support | Docs + email | Standard | Priority | SLA + named contact |
Add-ons
Buy more only where you need it.
Add-ons attach to Team, Business, and Enterprise. Most teams never need them.
| Add-on | What it does | Price |
|---|---|---|
| Extended audit retention Push retention past your plan default — 1 year, 3 years, or indefinite. | Push retention past your plan default — 1 year, 3 years, or indefinite. | From $99/mo |
| SIEM streaming Stream the redacted audit log to Splunk, Datadog, S3, or any SIEM, per connection. | Stream the redacted audit log to Splunk, Datadog, S3, or any SIEM, per connection. | Roadmap — contact us |
| Additional protected agents Raise your agent ceiling past the plan, in blocks. | Raise your agent ceiling past the plan, in blocks. | Metered — contact us |
| Additional guarded actions Raise your monthly action ceiling, in blocks. | Raise your monthly action ceiling, in blocks. | Metered — contact us |
| SSO seats (dashboard) SAML / OIDC seats for the humans who manage policy and approvals. | SAML / OIDC seats for the humans who manage policy and approvals. | Roadmap — Enterprise |
Extended retention and SIEM streaming are included or discounted on Enterprise. Add-on prices are launch pricing and may change before general availability.
FAQ
Pricing questions, answered straight.
Is there really a free tier?
Yes. The Developer tier is free — sign up for a hosted workspace, no credit card. You get the full policy / approval / signed-execution / audit envelope, 1 protected agent, 10,000 guarded actions a month, and 7-day retention. A real tier, not a 14-day trial that expires.
Why don’t you charge per seat?
Because seats don’t track risk. Adding a teammate to the dashboard doesn’t make your agents more dangerous — giving an agent production credentials does. We meter protected agents (machines that can act) and guarded actions (decisions on risky actions), so your bill scales with exposure, not headcount.
What’s a “guarded action”?
One allow / ask / block policy decision on something an agent tried to do — a refund, a delete, an email, a prod write. Each evaluation is one guarded action, including the ones Arc asks about or blocks. Stopping the dangerous ones is the work, so they count.
What’s the “$ guarded-spend pool”? Is it money I’m spending?
No. It’s not a wallet or a charge — it’s a plain-language measure of how much agent spend Arc is guarding for you each month (up to $250k/mo on Team). We don’t price the spend cap itself, because that’s a commodity in free gateways. You pay for the approval + signed-execution + audit envelope around it.
Isn’t a spend cap just what LiteLLM / Portkey / a gateway already does?
A gateway can cap a budget. It can’t pause a risky action for a human to approve, can’t hand your app a signed (ES256) request it can independently verify before executing, and can’t keep a redacted, hash-chained audit log that proves who approved what. Arc does the spend cap and the part gateways skip. Many teams run Arc alongside a gateway.
What happens if I go over my agent or action limit?
Nothing breaks. Limits are soft — Arc will not silently block your production traffic because you crossed a number on a pricing page. You’ll see usage in the dashboard and we’ll reach out to move you to the right tier (or add a block via an add-on). Your guardrails keep working the whole time.
Is billing self-serve today?
Yes. Sign up free, then upgrade to Team or Business in the console (Settings → Plan & billing) — self-serve Stripe Checkout, with the customer portal to change card, plan, or cancel anytime. Enterprise stays founder-led for single-tenant and custom terms.
Can I switch plans or cancel?
Yes. Move up or down as your agent footprint changes, and cancel anytime — paid plans are month-to-month with no lock-in. Annual pricing is available for Business and Enterprise on request.
Is Arc a replacement for Clerk or Auth0?
No. Clerk and Auth0 authenticate humans. Arc controls what AI agents are allowed to do once they’re acting — the policy, the approval, the signed execution, the audit. They’re complementary.
Is Arc production-ready?
Arc is early and shipping fast. The core is real and in use: agent registry, policy engine, approvals, signed execution, hash-chained audit, spend caps, the TypeScript SDK + CLI, and the MCP adapter. We’re honest about what’s still maturing — deeper RBAC, KMS-backed key management, external immutable export at scale — ask us and we’ll tell you exactly where a capability stands for your use case.
Arc
Put a cap on the bill and a gate on the damage.
Start free — sign up for a hosted workspace and protect your first agent on the Developer tier, no credit card. Move to a paid tier when your agents hit production.
The Developer tier is free — a hosted workspace and the @geostack/arc SDK. Team and Business are self-serve in the console (Stripe Checkout); Enterprise is founder-led. No per-seat surprises, no lock-in.