Sandbox
The sandbox is a hosted environment where you can evaluate OpenProdkt end to end with your own organization: write policies, call the real decision point from an SDK, handle approvals, and read the audit trail. It is for evaluation, not production use.
Access is by request. OpenProdkt reviews and approves every request before your organization exists.
How access works
Each step leads to the next in order. Approval is the only step that depends on OpenProdkt; everything after it is self-guided.
Step by step
1. Request access
Use the Register tab on the Console sign-in page, or call the API:
POST https://api.sandbox.openprodkt.com/auth/register
Content-Type: application/json
{
"organizationName": "Acme Ltd",
"email": "owner@acme.com",
"reason": "Evaluating runtime governance for our AI agents"
}
organizationName, email, and reason are all required. The response is 202 Accepted with a requestId and status: "pending". No organization or account exists yet, so you cannot sign in.
OpenProdkt reviews every request by hand. If you want your organization to start with sample data, say so in reason, for example "Prepopulate with sample data". See Sample data.
Check the status at any time. The endpoint needs no authentication:
GET https://api.sandbox.openprodkt.com/auth/onboarding-requests/{requestId}
| Status | Meaning |
|---|---|
pending | Waiting for review. You cannot request a sign-in code yet. |
approved | Your organization has been created. Continue with step 3. |
denied | The request was declined. The response includes the reason and the time after which you can request again. |
A second request from the same email is refused while one is pending, while a denial cool-down is active, and while you already have a trial.
2. Wait for approval
OpenProdkt reviews the request and sends you an email when it is approved. Approval creates your organization with you as its owner, in sandbox mode (see what you get). The 30-day trial starts at approval.
Sample data
If you asked for sample data, OpenProdkt fills your new organization right after approval with six example assets, six policies, 35 past decisions, five approvals in different states, five discovered assets, and one webhook endpoint. Use them to explore the Console, then change or delete them like your own data.
OpenProdkt can also reset a sandbox organization to this sample data later, if you ask. A reset:
- replaces all assets, groups, policies, decisions, approvals, discovered assets, webhook endpoints, and usage counters with the sample data;
- removes every API key and signs everyone out, so create a new API key after a reset;
- keeps you as the owner and removes any other members;
- keeps the audit trail, and adds an entry recording the reset.
If you sign in before sample data is added, you are signed out once and sign in again.
3. Sign in
Request a one-time code, then verify it. The code is sent to the email address you registered with.
POST https://api.sandbox.openprodkt.com/auth/request-code
Content-Type: application/json
{ "email": "owner@acme.com" }
POST https://api.sandbox.openprodkt.com/auth/verify
Content-Type: application/json
{ "email": "owner@acme.com", "code": "482916" }
The response contains accessToken. The Console does this for you. A session ends after 30 minutes without activity and after 12 hours in total. See Quickstart.
4. Create an API key
Applications authenticate with an API key, not with your sign-in token. Only the owner can create keys. Create one in the Console, or with the API:
POST https://api.sandbox.openprodkt.com/v1/api-keys
Authorization: Bearer <jwt>
Content-Type: application/json
{
"name": "evaluation",
"scopes": ["decision:run", "assets:read", "policies:read", "audit:read"],
"expiresAt": "2026-12-31T00:00:00Z"
}
- The key is shown once. Store it in a secret manager; OpenProdkt keeps only a hash.
expiresAtis required and may be at most 365 days away. Your trial still ends after 30 days.- Grant only the scopes the application needs.
decision:runis enough to ask for decisions. - API keys cannot create other API keys and cannot approve requests.
5. Define an asset and a policy
An asset is the thing you want to govern, such as an agent, tool, or API operation. A policy decides what happens when it is used. Create both in the Console as the owner, or follow Quickstart steps 6 to 8. For a first test, create:
- one asset
- one policy on it that returns
approval_requiredwhenmetadata.amountis greater than5000
A new organization starts empty. Until an asset exists and has an active policy, calls are not governed (see limitations).
6. Call the decision point from an SDK
Install an SDK from the SDK Manual, then point it at the sandbox. Keep the base URL configurable.
from openprodkt_governance import GovernanceClient, PendingApproval
client = GovernanceClient("https://api.sandbox.openprodkt.com", "<your-api-key>")
try:
decision = client.authorize(
asset_id="<asset-id>",
operation_id="evaluation-0001",
metadata={"operation": "payment.execute", "amount": 9000},
)
print(decision.decision, decision.decision_id)
except PendingApproval as pending:
print("waiting for approval", pending.approval_id, pending.status_url)
Use a new operation_id for each distinct operation and reuse it when you retry the same one. The Go and TypeScript SDKs follow the same pattern. See the SDK Manual.
Expected outcomes for the example above:
| Request | Outcome |
|---|---|
amount of 1000 | allow |
amount of 9000 | PendingApproval (HTTP 202) until the owner approves in the Console |
Retry after approval, same operation_id | allow, once. The approval is single-use |
7. Review the result
In the Console, open the decision history and audit views. With the API:
GET /v1/auditfor the audit trailGET /v1/usagefor consumption against your limitsGET /v1/approvalsfor pending and resolved approvals
The audit trail records the decision, the policy and reason, and a digest of the request. It does not store your request payload. Read Audit for what is and is not recorded.
8. Try the approval flow
The owner approves or rejects a pending request in the Console. Approvals need a signed-in user, not an API key, and the approver cannot be the principal that made the request. With one seat for the owner and two more available, you can invite a second user as an approver. See Approvals & Webhooks.
What you get
| Item | Sandbox value |
|---|---|
| Trial length | 30 days from approval |
| Requests per second | 25 per organization, shared by all keys |
| Decisions per month | 500,000 |
| Approvals per month | 500,000 |
| Notifications per month | 1,000,000 |
| Seats | 3 |
| API key lifetime | Up to 365 days, but unusable once the trial ends |
These are the sandbox profile values in Limits and Configuration. They apply to the organization, not to a key or user.
Functionality
Included in a sandbox organization:
- Decisions: the real decision point through the Go, Python, and TypeScript SDKs, the transparent proxy, and the MCP adapter.
- Assets and discovery: register assets, and observe unregistered ones that applications try to use, then register them.
- Policies: create, clone, activate, roll back, and test with the Console playground. See Policy Engine.
- Approvals and webhooks: approval workflow with email and webhook notification. Webhook endpoints must be public HTTPS addresses.
- Audit and usage: decision history, explanations, and usage counters.
- Groups and members: up to 3 seats.
- API keys: create, rotate, and revoke.
Not included:
- Evidence export and audit evidence packages.
- SSO, custom roles, and advanced AI features.
- Billing and plan changes. The sandbox is free and cannot be upgraded in place.
- Any uptime, availability, or performance commitment. See Performance Objectives for the hosted objectives, which are not a guarantee.
Limitations
- Shared evaluation environment. Other partners' organizations are isolated from yours, but the environment is shared and can change or restart. Do not put production data, real customer data, or secrets in it.
- Not governed until configured. A call for an asset that is not registered returns
allowwith reasonasset_not_registered. An asset with no active policy returnsallowwithno_policies_assigned, and one whose policies do not match returnsallowwithno_matching_policy. These allow-by-default outcomes apply to the sandbox as they do everywhere. Register assets and write policies before you rely on a result. - Rate and quota errors. Above 25 requests per second you receive HTTP
429; retry later. Above a monthly quota you receive HTTP402and nothing is evaluated. - Email dependent. Sign-in codes and approval notices are sent by email. Check spam folders if they do not arrive.
- Webhooks. Targets on loopback, private, or link-local addresses are rejected, so a local development server cannot receive webhooks directly. Use a public HTTPS endpoint.
- Audit protection. The audit trail is useful for evaluation, but it is not write-once evidence. See Audit for exactly which records are protected and when.
- No compliance claims. The sandbox is not certified or assessed against any regulatory or security standard.
When the trial ends
- Day 30: your license expires. API calls and Console requests are refused with HTTP
402and the errorsandbox trial expired. Nothing is evaluated. - At least seven days later (cleanup runs periodically): your organization is closed. All API keys and sessions are revoked, and the owner and member accounts are removed. You receive an email. Audit events are not deleted.
- Seven days after the closure you can request access again.
A new request starts from step 1 and needs a new approval. It creates a new, empty organization; policies, assets, and keys from the previous trial are not restored. Copy anything you want to keep, such as policy definitions, before the trial ends.
Troubleshooting
| Symptom | Likely cause |
|---|---|
register returns an error about a pending request or recent denial | A request is already open for that email, or a cool-down is active. Check the request status. |
register says the trial recently ended | You are inside the 7-day wait after a trial closed. |
| No sign-in code arrives | The request is not approved yet, the code expired, or the email is in spam. Codes can be requested again after a short wait. |
401 from the SDK | The key is wrong, revoked, or expired. |
403 from the SDK | The key lacks the scope for that call. |
402 with sandbox trial expired | The 30 days have passed. |
402 with decision_quota_exceeded | The monthly decision quota is used up. |
429 | You exceeded 25 requests per second. |
allow with asset_not_registered | The asset is not registered in your organization. |
For general errors, see Troubleshooting and Reason Codes.