canonical: https://jentic.com/apis/openhelm.ai/openhelm

# OpenHelm Public API

The OpenHelm Public API runs prompt-driven tasks on OpenHelm Cloud and returns structured, schema-shaped results asynchronously. You author jobs, reusable task templates whose prompts carry {{placeholders}}, or send an inline prompt, then poll the run or receive a callback with a result, a high, medium, or low confidence level, evidence such as a screenshot URL and matched text, and token usage with cost_usd. Task failures come back as a structured failure_reason such as captcha, auth_required, or timeout rather than an HTTP error. The same API also gives an organization email inboxes with sending, threads, full-text search, human approval holds, sending domains, and suppressions.

## For AI agents

Run prompt-based tasks on OpenHelm Cloud and read back schema-shaped results with confidence and screenshot evidence, and send and triage email from agent inboxes with human approval holds.

## Scope

Does not train models, host your own model weights, or provide raw chat completion. Use for running OpenHelm task jobs and its agent email inboxes only.

## Capabilities

- Run an inline prompt with {{placeholder}} inputs and an output_schema, then poll GET `/runs/{id}` or receive a callback for the result
- Author reusable jobs with a prompt, model, and input and output schemas, and run them with parameters and an idempotency_key
- Read each run's confidence level, screenshot and matched-text evidence, failure_reason, and token usage with cost_usd
- Pause or resume every scheduled run in the account, and cancel or force an individual run
- Create email inboxes with an autonomy mode of auto, approve_first, or notify, and send single messages or batches of up to 100
- Release or reject outbound messages held for human approval
- Search an organization's mail by full text and list threads, messages, and attachments

## Use cases

### Delegating structured tasks from your own agent

An AI agent working through Jentic hands a task to OpenHelm with POST /runs, sending a prompt, inputs for its {{placeholders}}, and an output_schema. The call returns 202 with a task_id, and GET `/runs/{id}` later returns a status of succeeded, failed, or needs_you along with the schema-shaped result, a confidence level, and evidence. Your own Jentic One instance injects the oh_live_ key at call time.

Example prompt: POST /runs with prompt 'Find the current price of {{product}} on {{site}}', inputs for product and site, and an output_schema with a price field, then poll GET `/runs/{id}` until status is succeeded and return result and confidence

### Reusable monitoring jobs

For checks that repeat, POST /jobs stores a named template with a project_id, a prompt, and schemas, and PATCH `/jobs/{id}` sets a schedule. Runs fired with POST `/jobs/{id}/runs` carry their own inputs and an idempotency_key, and each result includes evidence with a screenshot_url and matched_text so a person can confirm what was checked. POST `/scheduler/pause` stops every scheduled firing at once if something goes wrong.

Example prompt: Create a job with POST /jobs for project_id p_123 with a prompt and output_schema, then run it with POST `/jobs/{id}/runs` passing inputs and an idempotency_key

### Agent email with a human approval step

An inbox created with POST /inboxes and autonomy set to approve_first holds outbound agent mail instead of delivering it. GET /approvals lists the held messages, and POST `/approvals/{message_id}/approve` releases one, optionally with edits, while POST `/approvals/{message_id}/reject` discards it with a reason. Messages report sent, pending_approval, or scheduled, and suppressed recipients are listed on the response.

Example prompt: Send a reply with POST `/inboxes/{id}/messages` on an approve_first inbox, then list GET /approvals and approve the held message with POST `/approvals/{message_id}/approve`

### Triage and search across inboxes

Support or sales mail arriving at OpenHelm inboxes can be searched with GET /search by query and inbox_id, and grouped with GET /threads. GET /attachments returns the files on a message or thread, and GET `/attachments/{id}` gives a signed download URL, so an agent can find the right conversation and pull its documents.

Example prompt: Call GET /search with query 'invoice' and an inbox_id, open the top result with GET `/threads/{id}`, and fetch its attachments with GET /attachments

## Key endpoints

| Method | Path | Description |
| --- | --- | --- |
| POST | `/runs` | Run an inline prompt asynchronously |
| GET | `/runs/{id}` | Poll a task result |
| POST | `/jobs` | Create a task template |
| POST | `/jobs/{id}/runs` | Run a job with parameters |
| POST | `/scheduler/pause` | Pause all scheduled runs |
| POST | `/inboxes` | Create an inbox |
| POST | `/inboxes/{id}/messages` | Send from an inbox |
| POST | `/approvals/{message_id}/approve` | Release a held message |

## Key resources

- **Runs** — Asynchronous task executions returning status, result, confidence, evidence, failure_reason, and usage
- **Jobs** — Reusable task templates with a prompt, model, input and output schemas, and an optional schedule
- **Projects and agents** — Projects group jobs; agents (Goals) are standing missions with their own jobs
- **Inboxes and messages** — Email inboxes with autonomy modes, sending, threads, search, and attachments
- **Approvals** — Outbound messages held for a human to approve or reject
- **Domains, routes, and suppressions** — Sending domains with warm-up and DNS checks, forwarding routes, and do-not-send addresses

## AI readiness

This API is usable in Jentic One now. Its AI-readiness score against Jentic's framework shows where it stands today and where improvements would make it even easier for agents to use.

- **Score:** 45 / 100
- **Maturity:** Foundational
- **Dimensions:**
  - Foundational Compliance: 53 / 100
  - Developer Experience & Jentic Compatibility: 57 / 100
  - AI-Readiness & Agent Experience: 23 / 100
  - Agent Usability: 91 / 100
  - Security: 60 / 100
  - AI Discoverability: 69 / 100
- **View full report:** https://jentic.com/apis/openhelm.ai/openhelm/scorecard
- **How the score is calculated:** https://docs.jentic.com/reference/api-readiness-framework/overview/
- **More about the dimensions:** https://docs.jentic.com/reference/api-readiness-framework/specification/#dimensional-model-overview

### Score it yourself

Every API in the directory is allowlisted, so you can re-score it with no key required.

- **Score your own API:** https://jentic.com/scorecard.md
- **Scoring CLI agent skill:** https://github.com/jentic/jentic-api-scorecard/blob/main/skills/jentic-api-scorecard/SKILL.md

```sh
npx @jentic/api-scorecard-cli score <openapi-url>
```

## Why Jentic

- **Setup:** Wiring OpenHelm by hand means minting an oh_live_ key, sending it as a bearer token, polling asynchronous runs or hosting a callback, and handling 202 batch responses with per-item outcomes. Through Jentic you install once, import OpenHelm from the API Directory, store the key once, and your agent calls it.
- **Permission scoping:** OpenHelm puts job, run, and inbox ids in the URL path (`/jobs/{id}/runs`, `/inboxes/{id}/messages`), so a rule can pin your agent to one job or one inbox and the operations you pick. You choose the operations it may call, so account-wide ones like POST `/scheduler/pause` or DELETE `/connections/{id}` are not included unless you add them.
- **Credential handling:** Your OpenHelm key is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context.
- **Discovery method:** Agents search Jentic by intent such as 'run a task with structured output' or 'approve a held email', and Jentic returns the matching OpenHelm operation with its input schema, including prompt, inputs, and output_schema, so the agent calls it directly.

## Related APIs

- **Browser Use API** — Cloud browser sessions and automated browser tasks for AI agents
- **Browserbase API** — Headless browser sessions in the cloud for automation and scraping
- **Anjal API** — Email inboxes built for AI agents
- **OpenAI API** — Language model completions and embeddings

## FAQ

### What authentication does the OpenHelm Public API use?

The OpenHelm Public API uses a bearer token per its OpenAPI spec: send Authorization: Bearer with a key starting oh_live_, minted under Settings, API & Webhooks in OpenHelm. The spec applies it to every operation except GET /plans, which lists plans publicly with no key. Through Jentic the key is stored by your own instance and attached to each call.

### How do I get a run's result from the OpenHelm Public API?

POST /runs and POST `/jobs/{id}/runs` return 202 with a task_id and poll_url. Poll GET `/runs/{id}`, or pass a callback_url to receive a run.completed or run.failed event. The result object follows your output_schema and comes with a confidence of high, medium, or low, evidence with a screenshot_url valid for 7 days, and usage including cost_usd. A failed task still returns 200 with a failure_reason such as captcha, site_down, or not_found.

### What are the rate limits for the OpenHelm Public API?

The OpenAPI spec does not state numeric request limits. Run endpoints document a 429 for too many concurrent runs and a 402 for subscription, trial, or quota gates, and inboxes carry a daily_trigger_cap setting; batch sends stop at a daily-cap failure. See https://openhelm.ai for plan limits.

### Is there an OpenHelm MCP server?

You don't need an MCP server to give your agent OpenHelm. Jentic connects it directly from the API Directory: import it, store your credential once, and your agent calls the run, job, and inbox operations it needs on demand without loading another server's tool definitions into its context.

### How do I run an OpenHelm task through Jentic?

Search Jentic for 'run a task and return structured output', load the POST /runs operation, and execute it with a prompt, inputs, and an output_schema. Then load GET `/runs/{id}` and poll it with the returned task_id until the status is succeeded or failed. To run it on your own infrastructure, install Jentic One from its GitHub repo.

### Can I limit what my agent is allowed to do with the OpenHelm Public API?

Yes. For a monitoring assistant, write a rule that allows POST `/jobs/{id}/runs` for one job id and GET `/runs/{id}`, and the agent can trigger that job and read its results while job deletion, scheduler pause, connection changes, and email sending stay out of reach. Every call it makes is logged.