> ## Documentation Index > Fetch the complete documentation index at: https://help.descript.com/llms.txt > Use this file to discover all available pages before exploring further. # List agent models > List the currently available agent models and the aliases that resolve to them. The `model` parameter on [POST /jobs/agent](#operation/agentEditJob) accepts any value listed under `availableModels[].id` or `aliases[].id`. Aliases let you target the latest recommended model for a given tier without chasing version bumps — for example, passing `claude-opus` always routes to whichever Claude Opus version Descript currently recommends. The catalog changes as models launch and retire, so this endpoint's live response is the source of truth — the example below is an abridged illustration, not the full list. Cost tiers are coarse buckets — `low`, `medium`, `high` — useful for showing users a relative price/performance signal. Exact pricing is reported per job via the `ai_credits_used` field on [GET /jobs/{job_id}](#operation/getJob). When `model` is omitted on `POST /jobs/agent`, the request defaults to `auto`, which selects a recommended model for your account. `auto` is a `medium`-cost option. For an `auto` request, `result.resolved_model` on [GET /jobs/{job_id}](#operation/getJob) reports `auto`; for an explicit model or alias it reports the canonical id that ran. ## OpenAPI ````yaml GET /agent/models openapi: 3.0.0 info: description: Descript API documentation. version: '1.2' title: Descript API x-logo: url: assets/descript-logo.svg altText: Descript license: name: Proprietary url: https://www.descript.com/terms servers: - url: https://descriptapi.com/v1 security: [] tags: - name: Getting started description: "The Descript API lets you programmatically create projects, import media, and edit your projects — all without opening the app.\n\nTo learn more, visit [descript.com/api](https://descript.com/api).\n\n## Create an API token\n\n1. In Descript, open **Settings** and select **API tokens** from the sidebar. Then click **Create token**.\n\n![Navigate to Settings > API tokens and click Create token](assets/token1.png)\n\n2. Give your token a name and select the Drive it should be associated with. Click **Create token**.\n\n\"Name\n\n3. Copy your token and store it in a safe place. You won't be able to view it again. If you lose it, you'll need to generate a new one.\n\n\"Copy\n\n> **Warning:** Treat your API token like a password. Anyone with your token can make API requests on your behalf using your account permissions. Never share your token publicly or commit it to source control.\n\nInclude the token as a Bearer token in the `Authorization` header of your API requests.\n\n## Import media into a new project\n\nYou can create a new project, import media, and place the media into a composition all in one API request using the [import endpoint](#operation/importProjectMedia). This step also transcribes and processes the media so that it's ready for you or the agent to edit.\n\nTo import files, pass in public or pre-signed URIs. Currently, the API does not support uploading a file directly. To test the API with an example file, use the demo video included in the sample request below.\n\nImporting and processing is an asynchronous job, so the response payload will contain a `job_id` for you to [query the status of the job](#operation/getJob) and information about the newly created project. Note that `project_id` and `project_url` are returned immediately alongside `job_id`, but opening the project in Descript will not always show the API's processing state in real time. To prevent unintended changes, we recommend that you do not make any changes to the project until the job has stopped.\n\n\n\n**Request**\n\n```bash\ncurl -X POST https://descriptapi.com/v1/jobs/import/project_media \\\n -H \"Authorization: Bearer YOUR_API_TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\n \"project_name\": \"My First Video\",\n \"add_media\": {\n \"demo.mp4\": {\n \"url\": \"https://test-files.descriptapi.com/demo-video.mp4\"\n }\n },\n \"add_compositions\": [\n {\n \"name\": \"Demo Video\",\n \"clips\": [\n { \"media\": \"demo.mp4\" }\n ]\n }\n ]\n }'\n```\n\n**Response**\n\n```json\n{\n \"job_id\": \"project-media-import-9d635d5b\",\n \"drive_id\": \"c9c5c47e\",\n \"project_id\": \"e2f89ce6\",\n \"project_url\": \"https://web.descript.com/e2f89ce6\"\n}\n```\n\n## Check for import completion\n\nPoll the [job status endpoint](#operation/getJob) using the `job_id` from the last step to check whether the import job is finished processing. When it is, you'll see `job_state: \"stopped\"`. You can then check the `results` object to see its full results.\n\nYou can also pass in a `callback_url` as a part of the first import request, and we'll ping you when the job has stopped with the same response payload.\n\n**Request**\n\n```bash\ncurl https://descriptapi.com/v1/jobs/project-media-import-9d635d5b \\\n -H \"Authorization: Bearer YOUR_API_TOKEN\"\n```\n\n**Response**\n\n```json\n{\n \"job_id\": \"project-media-import-9d635d5b\",\n \"job_type\": \"import/project_media\",\n \"job_state\": \"stopped\",\n \"project_id\": \"e2f89ce6\",\n \"project_url\": \"https://web.descript.com/e2f89ce6\",\n \"result\": {\n \"status\": \"success\",\n \"media_status\": {\n \"main.mp4\": {\n \"status\": \"success\",\n \"duration_seconds\": 69.477006\n }\n },\n \"created_compositions\": [\n { \"id\": \"f8e5088a-4d53-4aab-9d4f-c6624b7d7622\", \"name\": \"Demo Video\" }\n ]\n }\n}\n```\n\n## Prompt for edits with Agent Underlord\n\nOnce your media is imported, you can use the [agent edit endpoint](#operation/agentEditJob) to prompt Underlord for edits, just as you would in the app. Because it's an API, conversation and follow up questions aren't practical. So we recommend framing your edits as a one-shot prompt with all the information the agent needs.\n\nEditing can take some time, so the response also returns a `job_id` that you can use to [check the status of the job](#operation/getJob). You can also pass in a `callback_url` as a part of an agent request, and we'll ping you when the job has stopped.\n\n**Request**\n\n```bash\ncurl -X POST https://descriptapi.com/v1/jobs/agent \\\n -H \"Authorization: Bearer YOUR_API_TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\n \"project_id\": \"e2f89ce6\",\n \"prompt\": \"Add studio sound and captions\"\n }'\n```\n\n**Response**\n\n```json\n{\n \"job_id\": \"project-agent-edit-e2f89ce6\",\n \"drive_id\": \"c9c5c47e\",\n \"project_id\": \"e2f89ce6\",\n \"project_url\": \"https://web.descript.com/e2f89ce6\"\n}\n```\n\n## Wait for the agent to complete its job\n\nPoll the [job status endpoint](#operation/getJob) using the `job_id`. When the agent job completes successfully, the response includes a summary of what it accomplished (see the `result.agent_response` field). Use the project URL to open the project in Descript and review Underlord's changes.\n\n**Request**\n\n```bash\ncurl https://descriptapi.com/v1/jobs/project-agent-edit-e2f89ce6 \\\n -H \"Authorization: Bearer YOUR_API_TOKEN\"\n```\n\nOnce the agent job is successfully complete, you’ll see a response from the agent with a brief summary of what it accomplished. You can then use the project url to review its changes directly in Descript.\n\n**Response**\n\n```json\n{\n\t\"job_id\":\"project-agent-edit-e2f89ce6\",\n\t\"job_type\":\"agent\",\n \"project_id\":\"YOUR_PROJECT_ID\",\n\t\"project_url\":\"https://web.descript.com/e2f89ce6\",\n\t\"job_state\":\"stopped\",\n\t\"created_at\":\"2026-02-09T05:42:27.554Z\",\n\t\"stopped_at\":\"2026-02-09T05:43:15.296Z\",\n\t\"drive_id\":\"1df135a5-dc4a-4dc3-8f7d-681cfbe961e4\",\n\t\"result\":{\n\t\t\"status\":\"success\",\n \"agent_response\":\"Done! I've applied Studio Sound to enhance your audio quality and added classic karaoke-style captions to your video.\",\n\t\t\"project_changed\":true,\n\t\t\"media_seconds_used\":0,\n\t\t\"ai_credits_used\":32\n }\n}\n```\n" - name: Using the CLI description: > The CLI wraps the API into a simple to use command line tool with interactive flows for setting up authentication, importing, and prompting the agent. It also has built-in polling for job completion. ## Requirements Before installing the CLI, you'll need **Node.js 24 or higher**. Visit [nodejs.org](https://nodejs.org/) to download and install the latest LTS version for your operating system. ## Install and set up First, install the latest version of the CLI using npm. ```bash npm install -g @descript/platform-cli@latest ``` Next, configure the CLI with your API key. ```bash descript-api config set api-key ``` ![Setting up the CLI](assets/cli-set-up.gif) ## Import media Use the `import` command to create a project by passing a project name and the link to any media you want to upload, or run `descript-api import` for interactive mode. The CLI shows live progress and outputs the project ID when done. ```bash descript-api import \ --name "My First Project" \ --media "https://test-files.descriptapi.com/demo-video.mp4" ``` ![Interactive CLI import](assets/cli-import.gif) ## Use the agent Use the `agent` command to edit a project by passing in the project id and an Underlord prompt. ```bash descript-api agent \ --project-id YOUR_PROJECT_ID \ --prompt "Remove filler words and add Studio Sound to all clips" ``` You can also ask Underlord create a new project from a prompt alone by writing the script for you! ```bash descript-api edit --new \ --prompt "Write a script about how to make great coffee" ``` ## All commands Run `descript-api help` to see the full list of available commands and options. - name: API Endpoints description: Import media, edit projects with AI, and query jobs and projects. - name: Direct file upload description: > Instead of providing a public URL, you can upload files directly from your local machine using the [import endpoint](#operation/importProjectMedia). The flow has three steps: request signed upload URLs, PUT your file bytes, then poll for completion. ## Step 1 — Request upload URLs Call the import endpoint with `content_type` and `file_size` instead of `url` for each media item you want to upload directly. **Request** ```bash curl -X POST https://descriptapi.com/v1/jobs/import/project_media \ -H "Authorization: Bearer YOUR_API_TOKEN" \ -H "Content-Type: application/json" \ -d '{ "project_name": "My Upload Project", "add_media": { "recording.mp4": { "content_type": "video/mp4", "file_size": 52428800 } }, "add_compositions": [ { "name": "Main", "clips": [ { "media": "recording.mp4" } ] } ] }' ``` The response includes an `upload_urls` object keyed by media reference ID. Each entry contains a signed `upload_url` (valid for 3 hours), plus `asset_id` and `artifact_id` for the created asset. **Response** ```json { "job_id": "project-media-import-a1b2c3d4", "drive_id": "c9c5c47e", "project_id": "e2f89ce6", "project_url": "https://web.descript.com/e2f89ce6", "upload_urls": { "recording.mp4": { "upload_url": "https://storage.googleapis.com/bucket/...", "asset_id": "d4e5f6a7-1234-5678-9abc-def012345678", "artifact_id": "a1b2c3d4-5678-9abc-def0-123456789abc" } } } ``` ## Step 2 — Upload the file PUT the raw file bytes to the signed URL. Use `Content-Type: application/octet-stream`. ```bash curl -X PUT \ -H "Content-Type: application/octet-stream" \ --data-binary @recording.mp4 \ "https://storage.googleapis.com/bucket/..." ``` The import job detects the upload automatically and begins processing. ## Step 3 — Poll for completion Check the job status the same way as a URL-based import — poll the [job status endpoint](#operation/getJob) with the `job_id`, or provide a `callback_url` in the original request. ```bash curl https://descriptapi.com/v1/jobs/project-media-import-a1b2c3d4 \ -H "Authorization: Bearer YOUR_API_TOKEN" ``` When the job reaches `job_state: "stopped"`, check `result.status` for success or failure. ## Mixing URL imports and direct uploads You can combine URL-based and direct upload media items in a single request. Items with `url` are fetched server-side; items with `content_type` and `file_size` return signed upload URLs. ```bash curl -X POST https://descriptapi.com/v1/jobs/import/project_media \ -H "Authorization: Bearer YOUR_API_TOKEN" \ -H "Content-Type: application/json" \ -d '{ "project_name": "Mixed Import", "add_media": { "intro.mp4": { "url": "https://example.com/intro.mp4" }, "recording.mp4": { "content_type": "video/mp4", "file_size": 52428800 } } }' ``` The response will include `upload_urls` only for the direct upload items. ## Required fields | Field | Type | Description | |-------|------|-------------| | `content_type` | string | MIME type of the file (e.g., `video/mp4`, `audio/wav`) | | `file_size` | integer | File size in bytes | | `language` | string | *(optional)* ISO 639-1 language code for transcription. Auto-detected if omitted. | - name: Authentication description: > The Descript API uses personal API tokens to authenticate requests. Tokens are scoped to a specific Drive and inherit your permissions on that Drive. To create a token, see the [Getting Started](#section/Getting-started/Create-an-API-token) guide. > **Warning:** Treat your API token like a password. Anyone with your token can make API requests on your behalf using your account permissions. Never share your token publicly or commit it to source control. ## Using your token Include the token as a Bearer token in the `Authorization` header of your API requests. **Example** ```bash curl -H "Authorization: Bearer YOUR_API_TOKEN" https://descriptapi.com/v1/status ``` - name: Rate Limiting description: > The Descript API implements rate limiting to ensure fair usage and protect service availability. When you exceed the rate limit, the API returns a `429 Too Many Requests` response. ## Rate Limit Headers When a rate limit is exceeded, the response includes the following headers: | Header | Description | |--------|-------------| | `Retry-After` | Number of seconds to wait before retrying the request | | `X-RateLimit-Remaining` | Number of requests remaining in the current window | | `X-RateLimit-Consumed` | Number of requests consumed in the current window | ## Handling Rate Limits When you receive a `429` response, use the `Retry-After` header to determine how long to wait before retrying. This approach is more efficient than using fixed delays or exponential backoff alone. - name: Edit in Descript description: > > **Note:** The Edit in Descript integration requires contacting Descript for access. [Reach out to us](https://descript.com/api) to get started. Edit in Descript API enables partners to give their users the ability to transfer audio or video content to Descript for editing. Edit in Descript buttons work by generating one-time use, public Import URLs to the Descript import UI that users can be automatically sent to. On that page, they can make a few simple selections before kicking off a Partner cloud storage to Descript cloud storage transfer. This will redirect them to a Descript Project ready for editing. Partners can initiate the request by securely sending an information schema backend-to-backend to the Descript API using a token, in exchange for the Import URL to redirect the user. Partners do not need to store this schema, as Descript will do so and use it to start fetching the files when the user confirms the action 1. When a user clicks `Edit in Descript`, partner's backend service makes POST request to: `https://descriptapi.com/v1/edit_in_descript/schema` with an authorization bearer token header and JSON schema body 2. Descript responds with either an Import URL or an error 3. Partner redirects the user's browser to the URL returned in step 2 or display an error message and link to help documentation ### Partner User Experience Some guidelines for partners as you consider this integration: * We recommend placing the `Edit in Descript` option next to your download options * If you offer multiple download options, such as combined vs. split audio/video files, we recommend placing this integration clearly in context with each option, or only the supported option, to help users understand what will be exported. * Each time you request an import link, a new one is generated. Import links expire after 3 hours. After using an import link, the only way to find an imported Project again is in Descript. * If an import link has expired or the contents of the schema has changed, please request a new import link with the updated schema. This will create a new Descript Project when used. * We will provide Descript-branded assets to fit your proposed placement of the `Edit in Descript` CTA and ask that you don't edit the assets beyond what we provide. We are happy to work with you on getting you the right assets for your placement. * Partners should provide error-handling for the POST request, at minimum displaying a generic error message and linking to a help article (we can provide a link for this if you prefer). * Progress will be conveyed to the user in the Descript side of the user experience. ### Descript User Experience When users are directed to a Descript Import URL, they'll be asked to either create an account or login in order to proceed. Next, they will be presented with a few options about how they'd like to import the data, such as where the new Descript Project should be created. They'll then be redirected to the Project, where they can monitor the progress of the import and start editing. - name: Export from Descript description: > Users of Descript currently have three options to export their edited content. They can export files in various formats, share a Descript link, or use our [one-click cloud export](https://help.descript.com/connect-with-us/hosting-partners) to publish directly to a partner. ### Roundtrip Metadata If Project data previously came from a partner via an Edit in Descript schema then any Descript Export pages will include `` tags which contains the `partner_drive_id` and `source_id` provided when originally importing into Descript. This allows partners to deduplicate data returning back to partner systems after editing in Descript. Both partner and source properties are included on all public Descript Export pages. ``` ``` paths: /agent/models: get: tags: - API Endpoints summary: List agent models description: > List the currently available agent models and the aliases that resolve to them. The `model` parameter on [POST /jobs/agent](#operation/agentEditJob) accepts any value listed under `availableModels[].id` or `aliases[].id`. Aliases let you target the latest recommended model for a given tier without chasing version bumps — for example, passing `claude-opus` always routes to whichever Claude Opus version Descript currently recommends. The catalog changes as models launch and retire, so this endpoint's live response is the source of truth — the example below is an abridged illustration, not the full list. Cost tiers are coarse buckets — `low`, `medium`, `high` — useful for showing users a relative price/performance signal. Exact pricing is reported per job via the `ai_credits_used` field on [GET /jobs/{job_id}](#operation/getJob). When `model` is omitted on `POST /jobs/agent`, the request defaults to `auto`, which selects a recommended model for your account. `auto` is a `medium`-cost option. For an `auto` request, `result.resolved_model` on [GET /jobs/{job_id}](#operation/getJob) reports `auto`; for an explicit model or alias it reports the canonical id that ran. operationId: listAgentModels responses: '200': description: Available agent models and aliases content: application/json: schema: type: object required: - availableModels - aliases properties: availableModels: type: array description: > Canonical model ids currently advertised by the public agent API, each tagged with a coarse cost tier. items: type: object required: - id - cost properties: id: type: string description: >- Canonical model id to pass as `model` on `POST /jobs/agent`. example: claude-opus-4.8 cost: type: string enum: - low - medium - high description: Relative cost tier for this model. example: high aliases: type: array description: > Friendly aliases that resolve to one of the `availableModels` at request time. Pass any alias `id` as `model` and the agent job result's `result.resolved_model` (on [GET /jobs/{job_id}](#operation/getJob)) will report the canonical id that actually ran. items: type: object required: - id - resolvesTo - description - cost properties: id: type: string description: Alias id callers can pass as `model`. example: claude-opus resolvesTo: type: string description: Canonical model id this alias currently resolves to. example: claude-opus-4.8 description: type: string description: Human-readable description of the alias's intent. example: Tracks stable Anthropic Claude Opus cost: type: string enum: - low - medium - high description: >- Relative cost tier of the model this alias resolves to. example: high examples: default_models: summary: Abridged example — call the endpoint for the current list value: availableModels: - id: auto cost: medium - id: claude-opus-4.8 cost: high - id: claude-haiku-4.5 cost: low aliases: - id: claude-opus resolvesTo: claude-opus-4.8 description: Tracks stable Anthropic Claude Opus cost: high '401': description: Unauthorized - missing or invalid authentication token content: application/json: schema: $ref: '#/components/schemas/Error401' '429': $ref: '#/components/responses/Error429Response' security: - bearerAuth: [] components: schemas: Error401: type: object properties: error: type: string example: unauthorized message: type: string example: Missing or invalid authentication token required: - error - message Error429: description: > Rate limit exceeded response. When this error is returned, the response includes headers to help you implement proper retry logic: - `Retry-After`: Number of seconds to wait before retrying - `X-RateLimit-Remaining`: Requests remaining in current window - `X-RateLimit-Consumed`: Requests consumed in current window type: object properties: error: type: string description: Error code indicating rate limit was exceeded example: rate_limit_exceeded message: type: string description: Human-readable error message example: Too many requests required: - error - message responses: Error429Response: description: | Too many requests - rate limit exceeded. Use the `Retry-After` header to determine when to retry. headers: Retry-After: description: Number of seconds to wait before retrying the request schema: type: integer example: 30 X-RateLimit-Remaining: description: Number of requests remaining in the current rate limit window schema: type: integer example: 0 X-RateLimit-Consumed: description: Number of requests consumed in the current rate limit window schema: type: integer example: 100 content: application/json: schema: $ref: '#/components/schemas/Error429' examples: rate_limit: summary: Rate limit exceeded value: error: rate_limit_exceeded message: Too many requests. Please try again later. securitySchemes: bearerAuth: type: http scheme: bearer description: >- Personal API token created in Descript Settings → API Tokens. See the Authentication section for details. ```` ## Related topics - [Track and understand your media minutes and AI credits](/billing-payments-plans/track-and-understand-your-media-minutes-and-ai-credits.md) - [Underlord (Beta): Your AI co-editor in Descript](/getting-started/underlord-beta-your-ai-co-editor-in-descript.md) - [Connect Descript to any AI assistant (custom MCP)](/api-and-mcp/mcp-custom.md) - [Connect Descript to Zapier](/api-and-mcp/zapier.md) - [Generative image and video models in Descript](/generative-media/models.md)