> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://developers.awork.com/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://developers.awork.com/_mcp/server. # Agents > Agents API Reference. Agents are AI teammates in awork. They can answer questions, prepare reports, and work on tasks and projects using the context of your workspace. Every user has a personal agent (the awork Agent), and workspaces can create custom agents with their own prompt, model configuration, capabilities, and connectors. ![The Agents home in awork](/_fern-img/9be79a1604c677528e2587d5014d634979dd7858d91c5133f64241dc4b4afca9.webp) ## Core Concepts * **Agents** are reusable configurations: a name, a system prompt, a model (or workspace preset), and a set of capabilities such as awork access, web search, or image generation. Changes to the system prompt are versioned and can be restored. * **Threads** are conversations with the awork Agent (`personal-agent`) or a custom agent (`custom-agent`). A thread can optionally be linked to a project, task, client, or document to give the agent context. * **Messages** are submitted into a thread as text, attachments, or both. A thread processes one message at a time — if the agent is still working, new messages are queued and processed in order. * **Connectors** link agents to external tools and data sources, and can be shared across the workspace. > **Note** > > Authentication works the same as for the rest of the awork API — see [Authentication](/authentication). All examples below use a Bearer token. > > Note that API keys authenticate as an API user rather than a person, and can therefore only interact with custom agents. The personal awork Agent (`personal-agent`) is tied to a real user, so `personal-agent` conversations require authenticating as that user via the [OAuth flow](/authentication). ## How to work with Agents ### Creating an agent To create a custom agent, provide a `name` and a model selection: either a workspace model preset (`presetKey`) or both `modelProvider` and `modelName` — but not both options at once. Typically you also set a `systemPrompt` that defines the agent's job: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents' \ -H 'Authorization: Bearer {token}' \ -H 'Content-Type: application/json' \ -d '{ "name": "Client Status Reporter", "description": "Prepares client-ready project status reports.", "systemPrompt": "Prepare concise client-ready project updates from awork data.", "presetKey": "balanced", "aworkAccessEnabled": true, "webSearchEnabled": true }' ``` The response contains the new agent, including its `id`, which you need to start threads for it: **`Response`** ```json title="Response" { "id": "23e68187-91d6-4f7e-8d31-e3f2c60d7427", "type": "custom", "name": "Client Status Reporter", "description": "Prepares client-ready project status reports.", "systemPrompt": "Prepare concise client-ready project updates from awork data.", "presetKey": "balanced", "createdOn": "2026-08-29T08:00:00Z", "createdBy": "9ad63972-4396-4b0b-9e7d-40b852cdebf8" } ``` To see the available models before creating an agent, use `GET /agents/models`. ### Getting default agent options Use `GET /agents/defaults` to retrieve the effective model, tool-access, connector, and skill options for a conversation. The `type` query parameter is required. In the awork UI, `personal-agent` is the **awork Agent** — the built-in personal agent available to the requesting user — while `custom-agent` refers to a custom agent configuration. For the awork Agent, pass `personal-agent` and omit `agentId`: **`Personal agent (awork Agent)`** ```sh title="Personal agent (awork Agent)" curl 'https://api.awork.com/api/v1/agents/defaults?type=personal-agent' \ -H 'Authorization: Bearer {token}' ``` For a custom agent, pass both `custom-agent` and that agent's id: **`Custom agent`** ```sh title="Custom agent" curl 'https://api.awork.com/api/v1/agents/defaults?type=custom-agent&agentId=23e68187-91d6-4f7e-8d31-e3f2c60d7427' \ -H 'Authorization: Bearer {token}' ``` `agentId` is required only when `type=custom-agent`; it is not used for `personal-agent`. If `type` is omitted or has any other value, the endpoint returns `400 Bad Request` with a validation error on `type`. Generated clients also expose `type` as a required, non-null argument. ### Copying an agent To duplicate an existing agent, use the dedicated copy endpoint. The source agent supplies the configuration and system prompt; you provide the new name and an optional description: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents/23e68187-91d6-4f7e-8d31-e3f2c60d7427/copy' \ -H 'Authorization: Bearer {token}' \ -H 'Content-Type: application/json' \ -d '{ "name": "Client Status Reporter (Marketing)", "description": "Status reports tailored to marketing projects." }' ``` The response is the new agent with its own `id`. You need read access to the source agent. ### Updating an agent and prompt history Update an agent's configuration with `PUT /agents/{agentId}`. System prompt changes are recorded as versions: * `GET /agents/{agentId}/systempromptversions` lists the saved prompt versions. * `POST /agents/{agentId}/systempromptversions/{systemPromptVersionId}/restore` restores a previous version. Agents can be archived and unarchived with `POST /agents/{agentId}/setarchived`. ### Starting a thread for an agent A thread is a conversation with one agent. The `type` field is required: use `custom-agent` together with an `agentId`, or `personal-agent` for the awork Agent (no `agentId` needed). Supply `content` (or attachments) to submit the first message right away — custom agent threads require a first message, while awork Agent threads can be created without one: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents/threads' \ -H 'Authorization: Bearer {token}' \ -H 'Content-Type: application/json' \ -d '{ "type": "custom-agent", "agentId": "23e68187-91d6-4f7e-8d31-e3f2c60d7427", "title": "Weekly client status report", "projectId": "c907b29e-f640-4639-910d-722623fd5b4a", "content": "Summarize this project'\''s progress, risks, and next steps." }' ``` The optional `projectId`, `taskId`, `clientId`, and `documentId` fields link the thread to an awork entity. The agent uses the linked entity as context for the whole conversation — for example, with a `projectId` set and awork access enabled, the agent can read the project's tasks and statuses to answer the question above; with a `documentId` set, it can read that doc's content. > **Visibility** > > Linking a thread to a project or task also changes who can see it. Threads without a project or task context are private to their creator. A project- or task-linked thread is visible to other users who can read that project or task **and** have the AI thread access permission (granted globally or through their project role) — for them it shows up in the project's agent threads, and they can read and contribute to the conversation. Deleting a thread remains possible only for its creator. **`Response`** ```json title="Response" { "id": "6cbb219e-839d-4d23-b273-3a78bbd842ae", "threadRunId": "8f934a03-8f8b-44ea-9137-3ca563bb75e8", "acceptedMessageId": "66fcbdc5-2ae6-43bd-b3ab-f18ccd24909a", "type": "custom-agent", "title": "Weekly client status report", "status": "accepted", "agentId": "23e68187-91d6-4f7e-8d31-e3f2c60d7427", "projectId": "c907b29e-f640-4639-910d-722623fd5b4a", "createdOn": "2026-08-29T08:00:00Z" } ``` Keep the `acceptedMessageId` — you use it to check the message's processing status. ### Sending messages to a thread Submit follow-up messages to an existing thread with text, attachments, or both. You can also override the model, reasoning level, tool access, connectors, and skills per message: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents/threads/6cbb219e-839d-4d23-b273-3a78bbd842ae/messages' \ -H 'Authorization: Bearer {token}' \ -H 'Content-Type: application/json' \ -d '{ "content": "Add the latest budget and launch-date risks to the report." }' ``` A thread processes one message at a time. The response confirms acceptance — it does not contain the agent's reply — and tells you whether the message started immediately or was queued: **`Response`** ```json title="Response" { "threadId": "6cbb219e-839d-4d23-b273-3a78bbd842ae", "threadRunId": "8f934a03-8f8b-44ea-9137-3ca563bb75e8", "acceptedMessageId": "66fcbdc5-2ae6-43bd-b3ab-f18ccd24909a", "isQueued": false, "status": "accepted" } ``` ### Adding context to a message There are two ways to give a specific message additional context: **Referencing awork entities inline.** Mention a doc, task, or project anywhere in the message `content` using entity-mention markup — the same format the awork editor produces. The agent resolves the reference with its awork tools and reads the entity's content: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents/threads/6cbb219e-839d-4d23-b273-3a78bbd842ae/messages' \ -H 'Authorization: Bearer {token}' \ -H 'Content-Type: application/json' \ -d '{ "content": "

Use the notes in for the report.

" }' ``` Supported `data-type` values include `document`, `task`, and `project`; `data-id` is the entity's id. The agent needs awork access enabled to resolve the reference. **Attaching files.** The `attachments` array accepts up to 20 files per message. Reference either an existing awork file the caller can read (`aworkFileId`) or a freshly uploaded temporary file (`temporaryFileId`, see [Files & Images](/files)) — exactly one of the two per attachment: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents/threads/6cbb219e-839d-4d23-b273-3a78bbd842ae/messages' \ -H 'Authorization: Bearer {token}' \ -H 'Content-Type: application/json' \ -d '{ "content": "Summarize the attached briefing.", "attachments": [ { "aworkFileId": "7a5c994a-320a-4850-a7a8-dad86809d3ef" } ] }' ``` Both patterns also work for the first message when creating a thread. To pin an entity as context for the whole conversation instead of a single message, use the thread-level `projectId`, `taskId`, `clientId`, or `documentId` fields when [starting the thread](#starting-a-thread-for-an-agent). ### Checking the processing status Poll the status of a submitted message using the server-assigned `acceptedMessageId`. The id stays the same while the message is queued and when it starts executing: **`Request`** ```sh title="Request" curl 'https://api.awork.com/api/v1/agents/threads/6cbb219e-839d-4d23-b273-3a78bbd842ae/messages/66fcbdc5-2ae6-43bd-b3ab-f18ccd24909a/status' \ -H 'Authorization: Bearer {token}' ``` **`Response`** ```json title="Response" { "messageId": "66fcbdc5-2ae6-43bd-b3ab-f18ccd24909a", "status": "completed", "hasAgentResponse": true, "responseId": "msg_01K4CLIENTSTATUS" } ``` Once completed, fetch the full conversation — including the agent's replies — with `GET /agents/threads/{threadId}/messages`. ### Cancelling a run If an agent is taking too long or is working on the wrong thing, cancel the active run. A successful cancellation also pauses dispatch of the next queued message, and the response returns the thread: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents/threads/6cbb219e-839d-4d23-b273-3a78bbd842ae/cancel' \ -H 'Authorization: Bearer {token}' ``` ## Files Agents and threads can both work with files: * **Agent files** are knowledge attached to an agent's configuration: upload via `POST /agents/{agentId}/files/byuploadid` (see [Files & Images](/files) for the upload flow), list with `GET /agents/{agentId}/files`, and delete with `DELETE /agents/{agentId}/files/{fileId}`. * **Thread files** are files produced in or attached to a conversation: retrieve them with `GET /agents/threads/{threadId}/files/{fileId}` and download their content via the `/download` sub-route. ## Connectors Connectors give agents access to external tools and data. List the connectors available in your workspace with `GET /agents/connectors` and the predefined presets with `GET /agents/connectors/presets`, then link one to an agent: **`Request`** ```sh title="Request" curl -X POST 'https://api.awork.com/api/v1/agents/23e68187-91d6-4f7e-8d31-e3f2c60d7427/connectors/6a1cf273-34a5-4dcd-90d6-6ba7728d488e' \ -H 'Authorization: Bearer {token}' ``` Unlink a connector with the corresponding `DELETE` request. Custom connectors can be made workspace-visible or private via `PUT /agents/connectors/{connectorId}/sharing`.