Infrastructure for the agentic economy

Your product,
clicked by humans.
called by agents.

AI agents are becoming the primary users of SaaS. Forge Box makes your product agent-ready. Two products. Signed receipts for every execution.

Get started free Sign in
$ npx forgebox init copy
execution_ledger — live
1REC  workflow refund_order · 14 steps
2GEN  openapi.json + policies + MCP tools
3DEP   live → api.example.com/v1/refund_order
4────────────────────────────────
5CALL agent claude-opus-4refund_order
6POL  spend_limit    PASS · approval_chain PASS
7EXEC 200 OK · 87ms · $0.002
8RCPT rcpt_9f2c41 · ed25519 ✓ signed

2
Products
12+
MCP Tools
$0.002
Per call
ed25519
Signed receipts

Two products, one platform

Pick one. Use both.
Either way, agents can use your product.

Every execution returns a signed receipt. No CAPTCHA. No human required.

AgentEndpoint

Workflow → API

Record any browser workflow. AgentEndpoint deploys it as a secure, versioned, auditable REST API that AI agents call directly.

  • Chrome extension records clicks, inputs, navigation
  • Auto-generates OpenAPI 3.1 manifest with policies
  • One-click deploy to production endpoint
  • YAML policy engine (allow/deny/flag per call)
  • Auto-generates MCP tools from workflows
  • Signed receipts (ed25519) for every execution
$0.002/call
Start with AgentEndpoint
AgentOS

Policy Sandbox

Agents test actions against a policy engine (risk-scored 0-100, allow/deny/flag) before executing in production. No code runs in Docker yet — that's coming in Phase F.

  • Docker isolation (seccomp, no-new-privileges)
  • Risk scoring 0-100 per action type
  • Stubbed APIs: Slack, GitHub, Stripe
  • Rule-by-rule policy evaluation trace
  • Self-serve sandbox keys — no sales call
  • Sandbox receipts with real signatures
$0.002/run
Start with AgentOS

AgentEndpoint

Record once.
Deploy everywhere.

No integration project. Record the workflow your team already does. We handle the rest.

01

Record

Chrome extension captures clicks, inputs, navigation.

02

Generate

OpenAPI manifest with policies, schemas, MCP tools.

03

Deploy

One click to a production endpoint. Versioned, auditable.

04

Enforce

YAML policy engine evaluates every agent call in real time.

05

Receipt

Signed receipt (ed25519). Every execution. Verifiable.


AgentOS

Test first.
Deploy confident.

Agents rehearse in an isolated sandbox before touching production.

01

Write

Agent writes code or selects a tool to test.

02

Isolate

Docker container with seccomp. Can't escape. Can't touch production.

03

Evaluate

Policy engine evaluates every action. PASS, DENY, or FLAG.

04

Score

Risk score 0-100. Shell = 80. API call = 15. Full transparency.

05

Receipt

Sandbox receipt with real signature. Promote when ready.


Comparison

AgentEndpoint vs AgentOS

Use together or independently. Both return signed receipts.

FeatureAgentEndpointAgentOS
Workflow recording✓ Chrome extension
OpenAPI manifest generation✓ Auto-generated
Policy engine✓ Real-time enforcement✓ Evaluation trace
Docker sandbox✓ Isolated containers
Risk scoring✓ 0-100 per action
Stubbed APIs✓ Slack, GitHub, Stripe
Signed receipts✓ ed25519✓ ed25519 (sandbox)
MCP tools✓ Auto-generated
Agent analytics
PricingFrom $0.002/callFrom $0.002/run

Pricing

Both products. One price.

AgentEndpoint and AgentOS included in every paid plan.

Free

$0/mo

For experimentation.

  • 1,000 calls/month
  • 100 sandbox runs/month
  • 3 workflows
  • Basic policies
  • Signed receipts
Start free
Pro — Recommended

$79/mo

For production workloads.

  • 50,000 calls/month
  • 10,000 sandbox runs/month
  • Unlimited workflows
  • Policy engine + MCP tools
  • Agent analytics
Start Pro

Business ($399/mo) and Enterprise (custom) available. Full refund within 30 days.


FAQ

Frequently asked questions

What does AgentEndpoint do?

Records your browser workflows (clicks, inputs, navigation) and deploys them as secure, versioned, auditable REST APIs. AI agents call these APIs directly. Includes OpenAPI manifest, YAML policy engine, MCP tools, and signed receipts (ed25519).

What does AgentOS do?

Policy engine that risk-scores every agent action 0-100 before it runs. Allow / deny / flag decisions based on spend limits, allowed domains, and rate limits. Today: regex-based risk classifier + stubbed tool responses for Slack, GitHub, Stripe. Tomorrow: real Docker isolation.

Do I need both?

No. AgentEndpoint works standalone — record workflows, deploy as APIs. AgentOS works standalone — test agent code in sandbox. Together they form a complete pipeline.

How much does it cost?

Both products included in all paid plans. Free: 1K calls/mo. Pro: $79/mo for 50K calls. Business: $399/mo for 500K calls. Overage: $0.002/call.

What is a signed receipt?

Every execution returns a JSON receipt with unique ID, timestamp, action, actor, policy decisions, cost, and an ed25519 signature. Verifiable. Retained for 7 years.

Can agents purchase without a human?

Yes. Checkout is completable entirely via API with no CAPTCHA. Agents can evaluate in sandbox, purchase, retrieve receipts, and request refunds autonomously.


Agent-readable by default

Agents read this page too.

Every claim on this site exists as a machine-readable artifact.

GET /.well-known/capabilities.json
{
  "organization": "Forge Box",
  "products": [
    {
      "name": "AgentEndpoint",
      "description": "Workflow → API",
      "capabilities": [
        "workflow.record",
        "manifest.generate",
        "endpoint.deploy",
        "policy.enforce",
        "receipt.verify"
      ],
      "pricing": { "per_call_usd": 0.002 }
    },
    {
      "name": "AgentOS",
      "description": "Policy sandbox",
      "capabilities": [
        "docker.sandbox",
        "risk.score",
        "stubbed.apis",
        "policy.evaluate"
      ],
      "pricing": { "per_run_usd": 0.002 }
    }
  ],
  "mcp": "https://mcp.forgebox.ai"
}