openapi: 3.2.0 info: title: Lab0 Knowledge API version: "1.0.0" summary: Upload documents and get answers grounded in them. description: | The Knowledge API stores your documents, makes them searchable, and answers questions using only what they say. This reference is generated from an OpenAPI document at `https://app.lab0.ai/app/api/knowledge/openapi.yaml`. 1. Upload documents into a **namespace**, for example one per end customer. 2. Wait until each file is `ready`. 3. Ask a question about a namespace. An agent searches and reads the documents, then answers and lists the files it retrieved passages from. ## Quickstart 1. Create an API key with **Read & write** access in Lab0 under **Account → API keys**. 2. Upload a document with [Upload a file](#tag/files/POST/api/knowledge/files). Namespaces are created on first use. 3. Poll [Get a file](#tag/files/GET/api/knowledge/files/{file_id}) every few seconds until its status is `ready` or `failed`. 4. Ask a question with [Ask a question](#tag/answers/POST/api/knowledge/respond). Every endpoint has code samples in curl, Python, Node.js and Go. [Ask a question](#tag/answers/POST/api/knowledge/respond) also has samples that stream the answer, under **Code Examples** in its language picker. ## Authentication Send a personal API key as a bearer token: `Authorization: Bearer lab0_sk_...`. A key acts as the Lab0 user who created it, and every request sees only that user's files. - **Read-only** keys can list and fetch files and namespaces. - **Read & write** keys can also upload, ask questions, and delete files and namespaces. Requests from a signed-in browser session also work. Uploads and questions sent that way must include the header `X-Lab0-Knowledge: 1`. ## Namespaces A namespace groups the documents a question can draw on. A question covers one namespace, optionally narrowed to some of its files. - Names are 1 to 128 ASCII letters, digits, `.`, `_`, `:` or `-`. Surrounding whitespace is removed. - When a request leaves the namespace out, it uses `default`. - A namespace is created by the first upload into it. It stays, even when empty, until you delete it. ## Processing An upload returns at once with the file in status `pending`. The file then moves through `extracting`, `structuring` and `indexing` to `ready`. Markdown files skip extraction and structuring. There is no row-count limit per table. Large tables are split into passages between rows, repeating headers and preserving source citations. A single exceptionally long row may be split to fit the embedding model's input. Long passages are read in pages of up to 4,000 Unicode characters. The agent can continue within a passage to reach evidence later in a long row. Supported ingestion formats: `.md`, `.pdf`, `.docx`, `.pptx`, `.xlsx`, `.xlsm`, `.xls`, `.csv`, `.tsv`, `.txt`, `.json`, `.xml`, `.rtf`, `.xbrl`, `.ixbrl`, `.html`, `.htm`, `.jpg`, `.jpeg`, `.png`, `.tif`, `.tiff`, `.bmp`, `.gif`, `.webp`. Convert legacy `.doc` and `.ppt` to `.docx` and `.pptx` before uploading. Unsupported formats may fail during extraction. If processing fails, the status becomes `failed` and `error` says which step failed, and why when the reason is known. Delete the file and upload it again to retry. When answering, the agent searches the namespace's index, which holds a file's passages as soon as they are written during `indexing`. It can read further through a document only once the file is `ready`. Wait for `ready` before relying on a file in answers. ## Answers Each question is answered on its own, with no conversation history. To ask a follow-up, include the context it needs in the prompt. - The agent can search and read only the namespace's documents, and does not search the web. It is instructed to answer only from them, and to say so when they don't cover the question rather than guess. - The answer takes whatever form the prompt asks for. The agent is told not to include images, and Markdown images and HTML `` tags are stripped. - `sources` lists the files the agent retrieved passages from. It can include a file the final answer did not use. - An answer usually takes several seconds and is cut off after three minutes. Set `stream: true` to receive the answer as server-sent events while it is written. The stream is: 1. one `meta` event 2. `chunk` events carrying text, and `tool` events while the agent searches and reads. A `chunk` with `"reset": true` means the text sent since the last tool call was not part of the answer: discard it. 3. one `done` event with the same body as the JSON response If answering fails mid-stream, an `error` event comes before `done`, and `done` carries no answer. Text received before an `error` event is not an answer: discard it. A stream that ends without `done` is incomplete. Lines starting with `:` are keep-alive comments. [Ask a question](#tag/answers/POST/api/knowledge/respond) shows an example stream. ## Limits | Limit | Default | | --- | --- | | File size | 50 MB | | Storage per user | 2 GB | | Files per namespace | 1,000 | | Namespaces per user | 100 | | Prompt length | 8,000 characters | | `file_ids` per question | 50 | | Question request body | 64 KB | A file over the size limit gets `413`. Uploads past the other storage limits get `409`: delete files or namespaces to make room. ## Errors Errors return a JSON body with a machine-readable `code`, a human-readable `message`, and a `request_id` to quote when reporting a problem. | Status | `code` | Meaning | | --- | --- | --- | | 400 | `invalid_input` | The request is malformed or a value is invalid. | | 401 | `unauthorized` | The API key is missing, invalid or expired. | | 403 | `unauthorized` | The key's scope doesn't allow this, or a browser request lacks `X-Lab0-Knowledge: 1`. | | 404 | `not_found` | The file or namespace doesn't exist or isn't yours. | | 409 | `conflict` | A storage limit would be exceeded. | | 413 | `invalid_input` | The file or request body is too large. | | 500 | `internal_error` | Something went wrong on Lab0's side. Retry, and quote `request_id` if it keeps happening. | | 502 | `provider_error` | The model or document search failed. Retry. | | 503 | `unavailable` or `auth_unavailable` | The service or sign-in is unavailable. Retry later. | | 504 | `unavailable` | The answer took longer than three minutes. Try a narrower question. | Any request can get `500` or `503`. A streamed answer that fails after the stream has started reports the failure in an `error` event instead of a `502` or `504`. servers: - url: https://app.lab0.ai/app description: Production security: - apiKey: [] tags: - name: Files description: Upload, inspect and delete documents. - name: Namespaces description: List and delete groups of documents. - name: Answers description: Ask questions answered from a namespace's documents. paths: /api/knowledge/files: post: operationId: uploadFile tags: [Files] summary: Upload a file description: | Stores the file and starts processing it in the background. The response is the file in status `pending`; poll [Get a file](#tag/files/GET/api/knowledge/files/{file_id}) until it is `ready` or `failed`. Requires a write key. parameters: - $ref: "#/components/parameters/BrowserGuard" requestBody: required: true content: multipart/form-data: schema: type: object required: [file] properties: file: type: string format: binary contentMediaType: application/octet-stream description: The document. Its file name is kept as the file's `name`. namespace: $ref: "#/components/schemas/NamespaceName" responses: "202": description: The file is stored and processing has started. content: application/json: schema: $ref: "#/components/schemas/File" example: id: kfile-3f9c2a7d1b6e4c08a5f2d9e1b7c4a6f0 namespace: "customer:4821" name: contract.pdf mime_type: application/pdf byte_size: 482113 status: pending created_at: "2026-10-09T09:30:00Z" updated_at: "2026-10-09T09:30:00Z" "400": $ref: "#/components/responses/InvalidInput" "401": $ref: "#/components/responses/Unauthorized" "403": $ref: "#/components/responses/Forbidden" "409": description: A storage limit would be exceeded. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: conflict message: This namespace already has the maximum of 1000 files. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e "413": description: The file is larger than the file size limit. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: invalid_input message: Files can be at most 50 MB. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e "503": $ref: "#/components/responses/Unavailable" get: operationId: listFiles tags: [Files] summary: List files description: Lists the files in one namespace, newest first. A namespace that doesn't exist has no files. parameters: - name: namespace in: query description: The namespace to list. Defaults to `default`. schema: $ref: "#/components/schemas/NamespaceName" responses: "200": description: The namespace's files. content: application/json: schema: type: object required: [files] properties: files: type: array items: $ref: "#/components/schemas/File" example: files: - id: kfile-3f9c2a7d1b6e4c08a5f2d9e1b7c4a6f0 namespace: "customer:4821" name: contract.pdf mime_type: application/pdf byte_size: 482113 status: ready created_at: "2026-10-09T09:30:00Z" updated_at: "2026-10-09T09:30:41Z" "400": $ref: "#/components/responses/InvalidInput" "401": $ref: "#/components/responses/Unauthorized" /api/knowledge/files/{file_id}: parameters: - $ref: "#/components/parameters/FileID" get: operationId: getFile tags: [Files] summary: Get a file description: Returns one file and its processing status. responses: "200": description: The file. content: application/json: schema: $ref: "#/components/schemas/File" examples: ready: summary: Ready value: id: kfile-3f9c2a7d1b6e4c08a5f2d9e1b7c4a6f0 namespace: "customer:4821" name: contract.pdf mime_type: application/pdf byte_size: 482113 status: ready created_at: "2026-10-09T09:30:00Z" updated_at: "2026-10-09T09:30:41Z" failed: summary: Processing failed value: id: kfile-9b2e7c4a1d3f4e6a8c0b5d7f9e1a3c5b namespace: "customer:4821" name: scan.pdf mime_type: application/pdf byte_size: 1048576 status: failed error: Document extraction failed. created_at: "2026-10-09T09:31:00Z" updated_at: "2026-10-09T09:31:12Z" "401": $ref: "#/components/responses/Unauthorized" "404": $ref: "#/components/responses/NotFound" delete: operationId: deleteFile tags: [Files] summary: Delete a file description: | Deletes the file, its stored content and its search index entries. A file can be deleted while it is processing; anything processing writes afterwards is removed too. Requires a write key. responses: "200": description: The deleted file, as it was when deleted. content: application/json: schema: $ref: "#/components/schemas/File" example: id: kfile-3f9c2a7d1b6e4c08a5f2d9e1b7c4a6f0 namespace: "customer:4821" name: contract.pdf mime_type: application/pdf byte_size: 482113 status: ready created_at: "2026-10-09T09:30:00Z" updated_at: "2026-10-09T09:30:41Z" "401": $ref: "#/components/responses/Unauthorized" "403": $ref: "#/components/responses/DeleteForbidden" "404": $ref: "#/components/responses/NotFound" /api/knowledge/namespaces: get: operationId: listNamespaces tags: [Namespaces] summary: List namespaces description: Lists your namespaces in name order, with their file counts and sizes. responses: "200": description: Your namespaces. content: application/json: schema: type: object required: [namespaces] properties: namespaces: type: array items: $ref: "#/components/schemas/Namespace" example: namespaces: - name: "customer:4821" file_count: 12 byte_size: 18340221 created_at: "2026-10-01T14:02:00Z" updated_at: "2026-10-09T09:30:00Z" "401": $ref: "#/components/responses/Unauthorized" /api/knowledge/namespaces/{namespace}: delete: operationId: deleteNamespace tags: [Namespaces] summary: Delete a namespace description: | Deletes the namespace with all its files, their stored content and the namespace's search index. Requires a write key. parameters: - name: namespace in: path required: true description: The namespace to delete. schema: $ref: "#/components/schemas/NamespaceName" responses: "200": description: The namespace is deleted. content: application/json: schema: type: object required: [deleted_files] properties: deleted_files: type: integer description: How many files were deleted with the namespace. example: deleted_files: 12 "400": $ref: "#/components/responses/InvalidInput" "401": $ref: "#/components/responses/Unauthorized" "403": $ref: "#/components/responses/DeleteForbidden" "404": $ref: "#/components/responses/NotFound" /api/knowledge/respond: post: operationId: respond x-codeSamples: - lang: curl label: "curl (streaming)" source: | curl -N https://app.lab0.ai/app/api/knowledge/respond \ --header "Authorization: Bearer $LAB0_API_KEY" \ --header "Content-Type: application/json" \ --data '{"prompt": "How much notice is needed to terminate for convenience?", "namespace": "customer:4821", "stream": true}' - lang: python label: "Python (streaming)" source: | import json import os import requests response = requests.post( "https://app.lab0.ai/app/api/knowledge/respond", headers={"Authorization": "Bearer " + os.environ["LAB0_API_KEY"]}, json={ "prompt": "How much notice is needed to terminate for convenience?", "namespace": "customer:4821", "stream": True, }, stream=True, ) if not response.ok: raise RuntimeError(f"{response.status_code}: {response.text}") response.encoding = "utf-8" event = None for line in response.iter_lines(decode_unicode=True, delimiter="\n"): if line.startswith("event: "): event = line.removeprefix("event: ") elif line.startswith("data: "): data = json.loads(line.removeprefix("data: ")) if event == "chunk": # A reset withdraws the text written since the last tool call. An # app would clear it; a terminal can't, so this starts a new line. print("\n" if data.get("reset") else data["text"], end="", flush=True) elif event == "done": print("\nSources:", [source["file_name"] for source in data["sources"]]) elif event == "error": raise RuntimeError(data["message"]) - lang: typescript label: "TypeScript (streaming)" source: | const response = await fetch("https://app.lab0.ai/app/api/knowledge/respond", { method: "POST", headers: { Authorization: `Bearer ${process.env.LAB0_API_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ prompt: "How much notice is needed to terminate for convenience?", namespace: "customer:4821", stream: true, }), }); if (!response.ok) { throw new Error(`${response.status}: ${await response.text()}`); } const reader = response.body!.pipeThrough(new TextDecoderStream()).getReader(); let buffer = ""; for (let read = await reader.read(); !read.done; read = await reader.read()) { buffer += read.value; let end: number; while ((end = buffer.indexOf("\n\n")) >= 0) { const block = buffer.slice(0, end); buffer = buffer.slice(end + 2); const event = /^event: (.*)$/m.exec(block)?.[1]; const data = /^data: (.*)$/m.exec(block)?.[1]; if (!event || !data) continue; // a keep-alive comment const payload = JSON.parse(data); if (event === "chunk") { // A reset withdraws the text written since the last tool call. An app // would clear it; a terminal can't, so this starts a new line. process.stdout.write(payload.reset ? "\n" : payload.text); } else if (event === "done") { console.log( "\nSources:", payload.sources.map((source: { file_name: string }) => source.file_name), ); } else if (event === "error") { throw new Error(payload.message); } } } - lang: go label: "Go (streaming)" source: | package main import ( "bufio" "bytes" "encoding/json" "fmt" "io" "net/http" "os" "strings" ) func main() { body, _ := json.Marshal(map[string]any{ "prompt": "How much notice is needed to terminate for convenience?", "namespace": "customer:4821", "stream": true, }) request, _ := http.NewRequest("POST", "https://app.lab0.ai/app/api/knowledge/respond", bytes.NewReader(body)) request.Header.Set("Authorization", "Bearer "+os.Getenv("LAB0_API_KEY")) request.Header.Set("Content-Type", "application/json") response, err := http.DefaultClient.Do(request) if err != nil { panic(err) } defer response.Body.Close() if response.StatusCode != http.StatusOK { message, _ := io.ReadAll(response.Body) panic(fmt.Sprintf("%s: %s", response.Status, message)) } scanner := bufio.NewScanner(response.Body) scanner.Buffer(make([]byte, 64<<10), 1<<20) // done carries the whole answer on one line var event string for scanner.Scan() { line := scanner.Text() if name, ok := strings.CutPrefix(line, "event: "); ok { event = name continue } data, ok := strings.CutPrefix(line, "data: ") if !ok { continue } var payload struct { Text string `json:"text"` Reset bool `json:"reset"` Message string `json:"message"` Sources []struct { FileName string `json:"file_name"` } `json:"sources"` } if err := json.Unmarshal([]byte(data), &payload); err != nil { panic(err) } switch { case event == "chunk" && payload.Reset: // A reset withdraws the text written since the last tool call. An // app would clear it; a terminal can't, so this starts a new line. fmt.Println() case event == "chunk": fmt.Print(payload.Text) case event == "done": fmt.Println() for _, source := range payload.Sources { fmt.Println("Source:", source.FileName) } case event == "error": panic(payload.Message) } } if err := scanner.Err(); err != nil { panic(err) } } tags: [Answers] summary: Ask a question description: | Answers the prompt from the documents in one namespace, optionally only from the files listed in `file_ids`. The response is JSON, or a stream of server-sent events when `stream` is true. Requires a write key. parameters: - $ref: "#/components/parameters/BrowserGuard" requestBody: required: true content: application/json: schema: $ref: "#/components/schemas/AnswerRequest" example: prompt: How much notice is needed to terminate for convenience? namespace: "customer:4821" responses: "200": description: The answer, as JSON or, with `stream` set, as server-sent events. content: application/json: schema: $ref: "#/components/schemas/Answer" example: answer: "Either party can terminate for convenience with **45 days'** written notice." sources: - file_id: kfile-3f9c2a7d1b6e4c08a5f2d9e1b7c4a6f0 file_name: contract.pdf model: openai/gpt-6-astra text/event-stream: itemSchema: $ref: "#/components/schemas/AnswerStreamEvent" examples: answer: summary: A streamed answer serializedValue: |+ event: meta data: {"namespace":"customer:4821","model":"openai/gpt-6-astra"} event: tool data: {"name":"search_documents","call_id":"call_1","status":"started"} event: tool data: {"name":"search_documents","call_id":"call_1","status":"completed"} event: chunk data: {"text":"Either party can terminate "} event: chunk data: {"text":"with **45 days'** written notice."} event: done data: {"answer":"Either party can terminate with **45 days'** written notice.","sources":[{"file_id":"kfile-3f9c2a7d1b6e4c08a5f2d9e1b7c4a6f0","file_name":"contract.pdf"}],"model":"openai/gpt-6-astra"} "400": $ref: "#/components/responses/InvalidInput" "401": $ref: "#/components/responses/Unauthorized" "403": $ref: "#/components/responses/Forbidden" "404": description: The namespace doesn't exist, or a file in `file_ids` isn't in it. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: not_found message: Knowledge namespace not found. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e "413": description: The request body is larger than 64 KB. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: invalid_input message: The request body is too large. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e "502": description: The model or the document search failed. Retry the request. A streamed answer reports this in an `error` event instead. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: provider_error message: Searching the documents failed. Try again. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e "503": $ref: "#/components/responses/Unavailable" "504": description: Answering took longer than three minutes. A streamed answer reports this in an `error` event instead. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: unavailable message: Answering took too long. Try a narrower question. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e components: securitySchemes: apiKey: type: http scheme: bearer description: A personal API key (`lab0_sk_...`), created in Lab0 under **Account → API keys**. parameters: FileID: name: file_id in: path required: true description: The file's ID. schema: type: string examples: [kfile-3f9c2a7d1b6e4c08a5f2d9e1b7c4a6f0] BrowserGuard: name: X-Lab0-Knowledge in: header description: Required, with the value `1`, on requests made with a browser session. API-key requests don't need it. schema: type: string const: "1" responses: InvalidInput: description: The request is malformed or a value is invalid. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: invalid_input message: "namespace must match ^[A-Za-z0-9._:-]{1,128}$" request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e Unauthorized: description: The API key is missing, invalid or expired. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: unauthorized message: Invalid or expired API key. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e Forbidden: description: The key's scope doesn't allow this, or a browser request lacks `X-Lab0-Knowledge`. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: unauthorized message: This API key is read-only. Create a key with write scope to perform this action. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e DeleteForbidden: description: The API key is read-only. Deleting needs a Read & write key. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: unauthorized message: API keys cannot perform destructive (DELETE) operations. Sign in to delete platform resources. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e NotFound: description: The resource doesn't exist or isn't yours. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: not_found message: Knowledge file not found. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e Unavailable: description: The service or sign-in is unavailable. Retry later. content: application/json: schema: $ref: "#/components/schemas/Error" example: code: auth_unavailable message: Sign-in is temporarily unavailable. request_id: Kq3vX9pLmT2wR8sYcN5bH7dJ4fG6aZ1e schemas: NamespaceName: type: string description: A namespace name. Defaults to `default` where it is optional. pattern: "^[A-Za-z0-9._:-]{1,128}$" examples: ["customer:4821"] File: type: object required: [id, namespace, name, mime_type, byte_size, status, created_at, updated_at] properties: id: type: string description: The file's ID. namespace: $ref: "#/components/schemas/NamespaceName" name: type: string description: The uploaded file's name. mime_type: type: string description: The file's media type, from the upload or its file name. byte_size: type: integer description: The file's size in bytes. status: type: string enum: [pending, extracting, structuring, indexing, ready, failed] description: Where the file is in processing. Wait for `ready` before relying on the file in answers. error: type: string description: Why processing failed. Present only when `status` is `failed`. created_at: type: string format: date-time updated_at: type: string format: date-time Namespace: type: object required: [name, file_count, byte_size, created_at, updated_at] properties: name: $ref: "#/components/schemas/NamespaceName" file_count: type: integer description: How many files the namespace holds. byte_size: type: integer description: The total size of its files in bytes. created_at: type: string format: date-time updated_at: type: string format: date-time AnswerRequest: type: object required: [prompt] properties: prompt: type: string minLength: 1 maxLength: 8000 description: The question or instruction. Surrounding whitespace is removed. namespace: $ref: "#/components/schemas/NamespaceName" file_ids: type: array maxItems: 50 items: type: string description: Answer only from these files, all of which must be in the namespace. Leave out to use every file. stream: type: boolean default: false description: Stream the answer as server-sent events. Answer: type: object required: [answer, sources, model] properties: answer: type: string description: The answer, in the form the prompt asked for. sources: type: array items: $ref: "#/components/schemas/Source" description: The files the agent retrieved passages from while answering. model: type: string description: The model that wrote the answer. Source: type: object required: [file_id, file_name] properties: file_id: type: string file_name: type: string AnswerStreamEvent: oneOf: - title: meta description: Sent first. type: object required: [event, data] properties: event: const: meta data: type: string description: JSON with the `namespace` being answered from and the `model` answering. contentMediaType: application/json contentSchema: $ref: "#/components/schemas/AnswerStreamMeta" - title: chunk description: Sent as the answer is written. type: object required: [event, data] properties: event: const: chunk data: type: string description: 'JSON with `text` to append to the answer, or `"reset": true` to discard the text since the last `tool` event.' contentMediaType: application/json contentSchema: $ref: "#/components/schemas/AnswerStreamChunk" - title: tool description: Sent when the agent starts or finishes searching or reading documents. type: object required: [event, data] properties: event: const: tool data: type: string description: JSON with the tool's `name` and `call_id`, and a `status` of `started`, `completed` or `error` (with `error`). contentMediaType: application/json contentSchema: $ref: "#/components/schemas/AnswerStreamTool" - title: error description: Sent when answering fails, just before `done`. Discard the text received before it. type: object required: [event, data] properties: event: const: error data: type: string description: JSON with a `message` saying why. contentMediaType: application/json contentSchema: $ref: "#/components/schemas/AnswerStreamError" - title: done description: Sent last. type: object required: [event, data] properties: event: const: done data: type: string description: JSON with the same body as the JSON response, or `{}` after an `error` event. contentMediaType: application/json contentSchema: oneOf: - $ref: "#/components/schemas/Answer" - type: object maxProperties: 0 AnswerStreamMeta: type: object description: The `data` of a `meta` event. required: [namespace, model] properties: namespace: $ref: "#/components/schemas/NamespaceName" model: type: string AnswerStreamChunk: type: object description: The `data` of a `chunk` event. required: [text] properties: text: type: string description: Text to append to the answer. Empty on a reset. reset: type: boolean description: When true, discard the text received since the last `tool` event. AnswerStreamTool: type: object description: The `data` of a `tool` event. required: [name, call_id, status] properties: name: type: string description: The tool, such as `search_documents`, `read_document` or `list_documents`. call_id: type: string description: Pairs a call's `started` and finishing events. status: type: string enum: [started, completed, error] error: type: string description: Why the call failed, when `status` is `error`. After a refused call, such as reading a file outside the namespace, the agent continues. If the documents couldn't be searched or read, answering fails with an `error` event. AnswerStreamError: type: object description: The `data` of an `error` event. required: [message] properties: message: type: string Error: type: object required: [code, message] properties: code: type: string description: A machine-readable error code, such as `invalid_input`, `unauthorized`, `not_found`, `conflict`, `provider_error` or `unavailable`. message: type: string description: A human-readable explanation. request_id: type: string description: Identifies the request when reporting a problem. details: type: object description: More about the error, for errors that have more to say.