> For the complete documentation index, see [llms.txt](https://agpt.co/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://agpt.co/docs/platform/api-and-integrations/api-guide/quickstart.md). # Quickstart Create an API key, add a sample agent, run it and read its output in about five minutes. Works the same on AutoGPT Cloud and self-hosted. In about five minutes you will create an API key, add a small sample agent to your library through the API, run it, and read its result. The sample agent uses no AI model and no third-party account, so it runs the same on AutoGPT Cloud and on a brand-new self-hosted instance. **You need** * An AutoGPT account on [AutoGPT Cloud](https://platform.agpt.co) or on your self-hosted instance. New accounts finish a short onboarding in the app before the settings pages open. * cURL, Python 3.9+ with `requests`, or Node.js 20+. * On AutoGPT Cloud only: an active plan or trial and a credit balance above zero, or runs are refused with `402 payment_required`. Self-hosted instances don't bill. {% hint style="info" %} Want your coding agent to do this for you? Paste the set-up prompt from [Build with AI coding agents](/docs/platform/api-and-integrations/api-guide/ai-coding-agents.md) into Claude Code, Codex, Cursor or Copilot. {% endhint %} {% stepper %} {% step %} ### Create an API key 1. In the AutoGPT app, open **Settings → AutoGPT API Keys**: [platform.agpt.co/settings/api-keys](https://platform.agpt.co/settings/api-keys) on Cloud, or `http://localhost:3000/settings/api-keys` on a self-hosted instance. 2. Click **Create Key** and name it `Quickstart`. 3. Tick these permissions: **Identity**, **Write Graph**, **Read Library**, **Write Library**, **Run Agent**, **Read Run**. (Write Library lets the full script at the end remove its sample agent again.) 4. Click **Create Key**, then copy the key. It starts with `agpt_` and is shown only once. Put the key and your instance's base URL in environment variables. Every example in these docs reads them: {% tabs %} {% tab title="AutoGPT Cloud" %} ```bash export AUTOGPT_API_URL="https://backend.agpt.co/api/v2" export AUTOGPT_API_KEY="agpt_..." ``` {% endtab %} {% tab title="Self-hosted (Docker Compose)" %} ```bash export AUTOGPT_API_URL="http://localhost:8006/api/v2" export AUTOGPT_API_KEY="agpt_..." ``` {% endtab %} {% tab title="Self-hosted (single container)" %} ```bash export AUTOGPT_API_URL="http://localhost:3000/_agpt/api/v2" export AUTOGPT_API_KEY="agpt_..." ``` {% endtab %} {% endtabs %} If your instance runs on another host or port, change the URL to match. [Cloud and self-hosted](/docs/platform/api-and-integrations/api-guide/environments.md) explains how to work it out. The cURL examples use Bash syntax; on Windows, run them in Git Bash or WSL. To use the Python or TypeScript scripts from PowerShell, set the variables with `$env:AUTOGPT_API_URL = "..."` and `$env:AUTOGPT_API_KEY = "agpt_..."`. In a project, keep both in a `.env` file that's in `.gitignore`: ```dotenv AUTOGPT_API_URL=https://backend.agpt.co/api/v2 AUTOGPT_API_KEY=agpt_... ``` {% endstep %} {% step %} ### Check the key ```bash curl -s "$AUTOGPT_API_URL/me" -H "X-API-Key: $AUTOGPT_API_KEY" ``` You should get your account back: ```json { "user_id": "2a9c2855-921d-421e-960b-dacb3a712a08", "email": "you@example.com", "name": "Ada", "timezone": "Europe/London", "organization": { "id": "5dad3c98-5c39-4782-b816-f9f6f2c53fe2", "name": "Ada", "is_personal": true }, "team": null, "scopes": ["IDENTITY", "WRITE_GRAPH", "READ_LIBRARY", "WRITE_LIBRARY", "RUN_AGENT", "READ_RUN"], "credential_type": "api_key" } ``` If you get something else, fix it before going on: | You got | It means | | ------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `401 unauthorized` | The key is wrong or revoked, or the header name is not `X-API-Key`. | | `403 forbidden` | The key lacks **Identity**, or you are no longer a member of the organization it was created in. The message says which. | | `404` | `AUTOGPT_API_URL` is wrong. It must end in `/api/v2`. | | `429` with `5 requests per 60s` | The API didn't recognise a key, so the request counted as anonymous: after five `401`s in a minute, you get this instead. Check the key and header, then wait a minute. | | A connection error | The instance isn't running at that address. | | {% endstep %} | | {% step %} ### Add a sample agent An agent is a graph of blocks: `nodes` are the blocks, and `links` connect one block's output to another's input. This one takes a name, puts it into a greeting with a text template, and returns the greeting. ```bash curl -s -X POST "$AUTOGPT_API_URL/graphs" \ -H "X-API-Key: $AUTOGPT_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "name": "Hello from the API", "description": "Greets whoever you name. Created by the AutoGPT API quickstart.", "nodes": [ {"id": "name-input", "block_id": "c0a8e994-ebf1-4a9c-a4d8-89d09c86741b", "input_default": {"name": "name", "title": "Your name", "value": "World"}}, {"id": "greeting-template", "block_id": "db7d8f02-2f44-4c55-ab7a-eae0941f0c30", "input_default": {"format": "Hello, {{ name }}!", "values": {}}}, {"id": "greeting-output", "block_id": "363ae599-353e-4804-937e-b2ee3cef3da4", "input_default": {"name": "greeting", "title": "Greeting"}} ], "links": [ {"id": "name-to-template", "source_id": "name-input", "source_name": "result", "sink_id": "greeting-template", "sink_name": "values_#_name"}, {"id": "template-to-output", "source_id": "greeting-template", "source_name": "output", "sink_id": "greeting-output", "sink_name": "value"} ] }' ``` The response is the saved graph, with an `id`, `"version": 1`, and the `input_schema` and `output_schema` the API worked out from the input and output blocks. Save the ID: ```bash export GRAPH_ID="992cb167-d6e3-4d7f-9dc3-002e34252de0" # the "id" from the response ``` Creating a graph also adds it to your library. You run agents through their **library agent**, which has its own ID: ```bash curl -s "$AUTOGPT_API_URL/graphs/$GRAPH_ID/library-agent" -H "X-API-Key: $AUTOGPT_API_KEY" ``` ```json { "id": "3acb5782-a2d9-4c3d-830a-6928317b280f", "graph_id": "992cb167-d6e3-4d7f-9dc3-002e34252de0", "graph_version": 1, "name": "Hello from the API", "input_schema": { "type": "object", "properties": { "name": { "title": "Your name", "default": "World" } }, "required": [] }, "output_schema": { "type": "object", "properties": { "greeting": { "title": "Greeting" } }, "required": ["greeting"] } } ``` The response has more fields than shown here. Save the library agent ID: ```bash export AGENT_ID="3acb5782-a2d9-4c3d-830a-6928317b280f" # the "id" from the response ``` {% endstep %} {% step %} ### Run it ```bash curl -s -X POST "$AUTOGPT_API_URL/library/agents/$AGENT_ID/runs" \ -H "X-API-Key: $AUTOGPT_API_KEY" \ -H "Content-Type: application/json" \ -H "Idempotency-Key: quickstart-$(date +%s)" \ -d '{"inputs": {"name": "Ada"}}' ``` `inputs` maps each property in the agent's `input_schema` to a value. The API answers `202 Accepted` straight away with the new run, before it starts: ```json { "id": "34cc713f-7b02-4622-95c7-5c8ea1a8b213", "graph_id": "992cb167-d6e3-4d7f-9dc3-002e34252de0", "graph_version": 1, "status": "QUEUED", "started_at": null, "ended_at": null, "inputs": { "name": "Ada" }, "cost_cents": 0, "duration_seconds": 0.0, "node_exec_count": 0 } ``` The `Idempotency-Key` makes the request safe to retry: sending it again with the same key returns this run instead of starting a second one. ```bash export RUN_ID="34cc713f-7b02-4622-95c7-5c8ea1a8b213" # the "id" from the response ``` {% endstep %} {% step %} ### Get the result Runs are asynchronous. Ask for the run until its `status` is `COMPLETED`, `FAILED` or `TERMINATED`: ```bash curl -s "$AUTOGPT_API_URL/runs/$RUN_ID" -H "X-API-Key: $AUTOGPT_API_KEY" ``` This agent finishes in a second or two: ```json { "id": "34cc713f-7b02-4622-95c7-5c8ea1a8b213", "status": "COMPLETED", "started_at": "2026-10-08T14:29:24.277000Z", "ended_at": "2026-10-08T14:29:25.098000Z", "inputs": { "name": "Ada" }, "outputs": { "greeting": ["Hello, Ada!"] }, "cost_cents": 0, "duration_seconds": 0.74, "node_exec_count": 3, "node_executions": [ "..." ] } ``` `outputs` maps each output name to a **list** of values, because an output block can fire more than once in a run. The greeting is `outputs["greeting"][0]`. {% endstep %} {% endstepper %} You have run an agent through the API. ## The whole flow as one script Each script creates the sample agent, runs it with an idempotency key, waits for it with a deadline, fails unless the output is exactly `{"greeting": ["Hello, Ada!"]}`, and removes the sample agent from your library again. The graph and its run stay on your account. If a request fails before the script has the agent's ID, remove the agent in the app. The scripts show the flow, not production code: they don't retry, and they don't branch on error codes. [Errors, rate limits, and pagination](/docs/platform/api-and-integrations/api-guide/api-conventions.md) covers both. {% tabs %} {% tab title="Python" %} {% code title="quickstart.py" lineNumbers="true" %} ```python """AutoGPT API quickstart: create a sample agent, run it, print its output, clean up. Needs: pip install requests (and python-dotenv to read a .env file) Env: AUTOGPT_API_URL, AUTOGPT_API_KEY (permissions: Identity, Write Graph, Read Library, Write Library, Run Agent, Read Run) """ import os import time import uuid import requests try: from dotenv import load_dotenv load_dotenv() except ImportError: pass API_URL = os.environ["AUTOGPT_API_URL"].rstrip("/") HEADERS = {"X-API-Key": os.environ["AUTOGPT_API_KEY"]} FINAL_STATUSES = {"COMPLETED", "FAILED", "TERMINATED"} HELLO_AGENT = { "name": "Hello from the API", "description": "Greets whoever you name. Created by the AutoGPT API quickstart.", "nodes": [ { "id": "name-input", "block_id": "c0a8e994-ebf1-4a9c-a4d8-89d09c86741b", # Agent Input "input_default": {"name": "name", "title": "Your name", "value": "World"}, }, { "id": "greeting-template", "block_id": "db7d8f02-2f44-4c55-ab7a-eae0941f0c30", # Fill Text Template "input_default": {"format": "Hello, {{ name }}!", "values": {}}, }, { "id": "greeting-output", "block_id": "363ae599-353e-4804-937e-b2ee3cef3da4", # Agent Output "input_default": {"name": "greeting", "title": "Greeting"}, }, ], "links": [ { "id": "name-to-template", "source_id": "name-input", "source_name": "result", "sink_id": "greeting-template", "sink_name": "values_#_name", }, { "id": "template-to-output", "source_id": "greeting-template", "source_name": "output", "sink_id": "greeting-output", "sink_name": "value", }, ], } def api(method: str, path: str, timeout: float = 30, **kwargs): response = requests.request( method, f"{API_URL}{path}", headers=HEADERS | kwargs.pop("headers", {}), timeout=timeout, **kwargs, ) if not response.ok: raise RuntimeError(f"{method} {path} -> {response.status_code}: {response.text}") return response.json() if response.content else None def wait_for_run(run_id: str, timeout_s: float = 300) -> dict: deadline = time.monotonic() + timeout_s delay, status = 1.0, "unknown" while (remaining := deadline - time.monotonic()) > 0: run = api("GET", f"/runs/{run_id}", timeout=min(30, remaining)) status = run["status"] if status in FINAL_STATUSES: return run if status == "REVIEW": raise RuntimeError(f"Run {run_id} is waiting for a human review") time.sleep(max(0, min(delay, deadline - time.monotonic()))) delay = min(delay * 1.5, 10) raise TimeoutError(f"Run {run_id} still {status} after {timeout_s}s") def node_errors(run: dict) -> list[str]: return [ message for node in run.get("node_executions") or [] if node["status"] == "FAILED" for message in node["outputs"].get("error", []) ] me = api("GET", "/me") print(f"Authenticated as {me['email']} in {me['organization']['name']}") graph = api("POST", "/graphs", json=HELLO_AGENT) agent = None try: agent = api("GET", f"/graphs/{graph['id']}/library-agent") print(f"Created agent {agent['id']}") run = api( "POST", f"/library/agents/{agent['id']}/runs", json={"inputs": {"name": "Ada"}}, headers={"Idempotency-Key": str(uuid.uuid4())}, ) print(f"Started run {run['id']} ({run['status']})") run = wait_for_run(run["id"]) if run["status"] != "COMPLETED" or run["outputs"] != {"greeting": ["Hello, Ada!"]}: raise RuntimeError(f"Run {run['id']} {run['status']}: {node_errors(run) or run['outputs']}") print(f"Run {run['status']}: {run['outputs']['greeting'][0]}") finally: if agent: api("DELETE", f"/library/agents/{agent['id']}") print("Removed the sample agent from the library") ``` {% endcode %} Run it with `python quickstart.py`. With `python-dotenv` installed it also reads a `.env` file in the current folder. It prints: ``` Authenticated as you@example.com in Ada Created agent 8f21cc9c-7345-4bc3-acea-125c4588fd2a Started run d4436ae4-b15f-4dd3-aa9c-66996befd0fb (QUEUED) Run COMPLETED: Hello, Ada! Removed the sample agent from the library ``` {% endtab %} {% tab title="TypeScript" %} {% code title="quickstart.mts" lineNumbers="true" %} ```typescript // AutoGPT API quickstart: create a sample agent, run it, print its output, clean up. // Needs: Node.js 20+. Run with: npx tsx quickstart.mts // Env: AUTOGPT_API_URL, AUTOGPT_API_KEY (permissions: Identity, Write Graph, // Read Library, Write Library, Run Agent, Read Run) const API_URL = process.env.AUTOGPT_API_URL!.replace(/\/$/, ""); const API_KEY = process.env.AUTOGPT_API_KEY!; const FINAL_STATUSES = new Set(["COMPLETED", "FAILED", "TERMINATED"]); const HELLO_AGENT = { name: "Hello from the API", description: "Greets whoever you name. Created by the AutoGPT API quickstart.", nodes: [ { id: "name-input", block_id: "c0a8e994-ebf1-4a9c-a4d8-89d09c86741b", // Agent Input input_default: { name: "name", title: "Your name", value: "World" }, }, { id: "greeting-template", block_id: "db7d8f02-2f44-4c55-ab7a-eae0941f0c30", // Fill Text Template input_default: { format: "Hello, {{ name }}!", values: {} }, }, { id: "greeting-output", block_id: "363ae599-353e-4804-937e-b2ee3cef3da4", // Agent Output input_default: { name: "greeting", title: "Greeting" }, }, ], links: [ { id: "name-to-template", source_id: "name-input", source_name: "result", sink_id: "greeting-template", sink_name: "values_#_name", }, { id: "template-to-output", source_id: "greeting-template", source_name: "output", sink_id: "greeting-output", sink_name: "value", }, ], }; async function api( method: string, path: string, body?: unknown, headers: Record = {}, timeoutMs = 30_000, ) { const response = await fetch(`${API_URL}${path}`, { method, headers: { "X-API-Key": API_KEY, "Content-Type": "application/json", ...headers }, body: body === undefined ? undefined : JSON.stringify(body), signal: AbortSignal.timeout(timeoutMs), }); if (!response.ok) { throw new Error(`${method} ${path} -> ${response.status}: ${await response.text()}`); } return response.status === 204 ? null : response.json(); } async function waitForRun(runId: string, timeoutMs = 300_000) { const deadline = Date.now() + timeoutMs; let delay = 1000; let status = "unknown"; for (let remaining = timeoutMs; remaining > 0; remaining = deadline - Date.now()) { const run = await api("GET", `/runs/${runId}`, undefined, {}, Math.min(30_000, remaining)); status = run.status; if (FINAL_STATUSES.has(status)) return run; if (status === "REVIEW") throw new Error(`Run ${runId} is waiting for a human review`); const pause = Math.max(0, Math.min(delay, deadline - Date.now())); await new Promise((resolve) => setTimeout(resolve, pause)); delay = Math.min(delay * 1.5, 10_000); } throw new Error(`Run ${runId} still ${status} after ${timeoutMs} ms`); } function nodeErrors(run: { node_executions?: { status: string; outputs: Record }[] }) { return (run.node_executions ?? []) .filter((node) => node.status === "FAILED") .flatMap((node) => node.outputs.error ?? []); } const me = await api("GET", "/me"); console.log(`Authenticated as ${me.email} in ${me.organization.name}`); const graph = await api("POST", "/graphs", HELLO_AGENT); let agent: { id: string } | undefined; try { agent = await api("GET", `/graphs/${graph.id}/library-agent`); console.log(`Created agent ${agent!.id}`); let run = await api( "POST", `/library/agents/${agent!.id}/runs`, { inputs: { name: "Ada" } }, { "Idempotency-Key": crypto.randomUUID() }, ); console.log(`Started run ${run.id} (${run.status})`); run = await waitForRun(run.id); const expected = JSON.stringify({ greeting: ["Hello, Ada!"] }); if (run.status !== "COMPLETED" || JSON.stringify(run.outputs) !== expected) { const errors = nodeErrors(run); throw new Error(`Run ${run.id} ${run.status}: ${JSON.stringify(errors.length ? errors : run.outputs)}`); } console.log(`Run ${run.status}: ${run.outputs.greeting[0]}`); } finally { if (agent) { await api("DELETE", `/library/agents/${agent.id}`); console.log("Removed the sample agent from the library"); } } ``` {% endcode %} Save it as `quickstart.mts` (the `.mts` extension lets it use top-level `await`) and run `npx tsx quickstart.mts`. On Node.js 23.6 or later you can run it directly, and load a `.env` file at the same time: `node --env-file=.env quickstart.mts`. It prints: ``` Authenticated as you@example.com in Ada Created agent 01322f04-808c-40ba-af2f-a71ac8133d6e Started run fd13103e-6caa-428b-9c1e-18196d0ff7f2 (QUEUED) Run COMPLETED: Hello, Ada! Removed the sample agent from the library ``` {% endtab %} {% endtabs %} The step-by-step version above leaves the sample agent in your library. Remove it in the app, or with `DELETE /library/agents/{agent_id}`. ## Next steps * **Run your own agents.** List your library with `GET /library/agents` (needs **Read Library**), read an agent's `input_schema` from `GET /library/agents/{agent_id}`, and run it the same way. Add **Read Integrations** to the key to check which credentials an agent needs. [Run agents](/docs/platform/api-and-integrations/api-guide/running-agents.md) covers credentials, files, human reviews, schedules and failures. * **Build agents in code.** [Build agents](/docs/platform/api-and-integrations/api-guide/building-agents.md) explains blocks, links and versions. * **Handle errors and limits.** [Errors, rate limits, and pagination](/docs/platform/api-and-integrations/api-guide/api-conventions.md). * **Connect an AI tool.** [MCP server](/docs/platform/api-and-integrations/api-guide/mcp-server.md) lets Claude, Cursor and other MCP clients do all of this conversationally. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://agpt.co/docs/platform/api-and-integrations/api-guide/quickstart.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.