> ## Documentation Index > Fetch the complete documentation index at: https://docs.getoneprofile.ai/llms.txt > Use this file to discover all available pages before exploring further. # Generate a reply draft, streaming the work > Draft with AI as a stream of Server-Sent Events. Writes exactly the reply the plain draft route writes, and says what it did on the way: a `step` frame for each piece of work once it is done (reading the conversation, reading the contact's LinkedIn profile when the conversation's agent does, writing the reply), each with its measured duration, then one `result` frame whose `data` is the draft, or null when nothing usable was written. Failures arrive as an `error` frame, because the stream has already answered 200: reason `not_found` when the conversation is gone, `limit` when the workspace has used today's AI Assistant allowance, `paused` when the AI Assistant is paused, and no reason for anything else, including a sequence with no audience table, whose sentence says so. Closing the connection does not stop the draft: it is still written and saved. ## OpenAPI ````yaml /openapi.json post /inbox-ai/threads/{thread_id}/draft/stream/ openapi: 3.1.0 info: contact: email: support@getoneprofile.ai name: Oneprofile Support description: >- REST API reference for Oneprofile, the GTM data layer for your CRM. Manage tables, enrichments, and integrations. Authenticate with an API key to get started. termsOfService: https://www.getoneprofile.ai/terms title: API Reference | Oneprofile version: '0.0' servers: - url: https://api.getoneprofile.ai security: [] tags: - description: >- The docked AI Assistant: its general chat about the workspace and the product, streamed as Server-Sent Events, and the one list of a person's threads across every assistant, with pin, rename, delete and the search-to-table link. name: assistant - description: >- The messages a person keeps to send the assistant again: the ones the workspace saved and the ones that ship with the product, with save, rename, edit and delete. name: assistant-prompt - description: >- Granted workspace actions: what the AI Assistant proposed or applied in a conversation, the approve and dismiss decisions on each card, the workspace's approval threshold and default grants, and the eight verbs. name: workspace-actions - description: >- The Getting started checklist: a workspace's progress through the six steps from its first people search to its first launched sequence. name: onboarding - description: Senders, connected mailboxes, tags, and the Google connect flow. name: senders - description: The workspace do-not-contact list, enforced before every send. name: suppression - description: >- The workspace's recontact rule: how long before anyone hears from us again. name: contact_cooldown - description: 'Sequence authoring: the graph, the table-backed audience, previews.' name: sequences - description: >- The outbound agent behind a sequence: the voice it writes in, its per-message prompts, what it wrote and what people told it. name: outbound-agents - description: >- The LinkedIn channel: connecting an account to a sender, its safety limits and history, and the protocol the browser extension speaks. name: linkedin - description: >- The approval queue: everything waiting on a person across every sender and channel, and the workspace's manual or automatic modes per channel. name: outreach - description: 'The unified outreach inbox: threads, replies, labels, events.' name: inbox - description: 'The AI reply layer: workspace settings, drafts, and approval.' name: inbox-ai - description: Notification rules, per-user preferences, and the Slack connection. name: notifications - description: Endpoints for seeing all available integration types. name: integration-types - description: Endpoints for creating and managing integrations. name: integrations - description: Endpoints for the generic OAuth install/claim app-listings engine. name: app-listings - description: Endpoints for creating and managing API keys. name: apikey - description: >- The Model Context Protocol endpoint: use Oneprofile from Claude Code, Claude Desktop, Cursor or any other MCP client, with an API key as the bearer token. name: mcp - description: Endpoints for managing billing and subscription information. name: billing - description: Endpoints for creating and managing tables. name: tables - description: Endpoints for table columns, formula previews, and on-demand column runs. name: table-columns - description: Endpoints for table rows and manual cell edits. name: table-rows - description: Endpoints for observing and cancelling table recompute runs. name: table-runs - description: >- Endpoints for org-shared per-column display preferences: hidden, pinned, width, and colour tint. name: table-ui-prefs - description: >- Endpoints for the org-shared names given to the column tint palette: what this organisation calls each colour. Org-level, not per-table. name: table-column-color-labels - description: >- Endpoints for the enrichment action catalog, model list, presets, draft generation, and saved agents. name: table-enrichments - description: >- Endpoints for teaching a saved agent from example rows: training sessions, per-row feedback, and the proposed prompt to approve. name: agent-training - description: >- The durable audit trail of a table's enrichment actions: configured, edited, run, cancelled, and settled cost. name: table-enrichment-audit-logs - description: Endpoints for saving and applying column templates. name: table-templates - description: Server-sent change feed for a table's grid. name: table-events - description: 'Endpoints for org-shared saved table views: sort, filters, hidden columns.' name: table-views - description: >- Endpoints for PER-USER table preferences: favourites and recently opened. Unlike the rest of the table surface these are personal, so two members of one organisation legitimately see different answers. name: table-user-prefs - description: >- Endpoints for building a table from an uploaded CSV: upload and preview the inferred columns, commit the reviewed mapping, then poll the row load. name: table-imports - description: Endpoint for downloading a table's rows as CSV. name: table-export - description: >- Endpoints for the integration a table imports from: attach and preview it, promote imported fields into columns, and run the import on demand. name: table-sources - description: >- Endpoints for a table's recurring import cadence: set it, pause it, and read the next fire times. name: table-schedules - description: >- Endpoints for pushing a table's rows out to a connected integration: list writable destinations, derive their write capabilities, and map export columns. name: table-writeback - description: >- Endpoints backing a lookup column's setup: the connections whose records a column could read, and the identity fields one record type can be looked up by. name: table-integration-lookup - description: >- Endpoints for sourcing leads from the lead database: count and preview matches, land them in a table, and keep saved searches. Alongside people and company searches, leads can be sourced from buying signals — what a company or person just did — with a catalogue endpoint behind the picker that builds them. name: find-leads - description: >- Endpoints for discovering a domain's subdomains from public certificate logs and landing the results in a table. name: find-subdomains - description: Endpoints for the credit balance meter and usage reporting. name: credits - description: >- Endpoints for the workspace referral program: the shareable link, the friends who joined through it, and claiming a code after signup. name: referral - description: >- Endpoints for the affiliate program's own dashboard: enrolling and claiming a code, the summary, stats and activity feed, the commission ledger, and identity verification and payouts. name: affiliate - description: >- Endpoints for the design partner program's customer surface: whether this workspace is a design partner, the badges its feedback has earned, and sending and reading that feedback. name: design-partner - description: >- Endpoints for the org-level AI business context — company description, ideal customer profile, buyer personas, interest signals and social proof — plus domain-based generation and the reference documents whose summaries ride alongside those fields. name: ai-context - description: >- Endpoints for the signed-in user's Google Drive connection: starting and finishing the OAuth grant, checking or dropping it, and minting the short-lived token the Google Picker needs. name: google-drive paths: /inbox-ai/threads/{thread_id}/draft/stream/: post: tags: - inbox-ai summary: Generate a reply draft, streaming the work description: >- Draft with AI as a stream of Server-Sent Events. Writes exactly the reply the plain draft route writes, and says what it did on the way: a `step` frame for each piece of work once it is done (reading the conversation, reading the contact's LinkedIn profile when the conversation's agent does, writing the reply), each with its measured duration, then one `result` frame whose `data` is the draft, or null when nothing usable was written. Failures arrive as an `error` frame, because the stream has already answered 200: reason `not_found` when the conversation is gone, `limit` when the workspace has used today's AI Assistant allowance, `paused` when the AI Assistant is paused, and no reason for anything else, including a sequence with no audience table, whose sentence says so. Closing the connection does not stop the draft: it is still written and saved. operationId: stream_generate_draft parameters: - in: path name: thread_id required: true schema: format: uuid title: Thread Id type: string requestBody: content: application/json: schema: $ref: '#/components/schemas/GenerateDraftSchema' required: true responses: '200': content: text/event-stream: schema: $ref: '#/components/schemas/AiDraftReadSchema' type: string description: >- A stream of Server-Sent Events; the final `result` frame's `data` is the draft, or null. '401': content: text/event-stream: schema: $ref: '#/components/schemas/UnauthorizedResponse' description: Unauthorized '403': content: text/event-stream: schema: $ref: '#/components/schemas/ForbiddenResponse' description: Forbidden '422': content: text/event-stream: schema: $ref: '#/components/schemas/UnprocessableEntityResponse' description: Invalid request made to the API. '500': content: text/event-stream: schema: $ref: '#/components/schemas/ServerErrorResponse' description: Internal server error. '502': content: text/event-stream: schema: $ref: '#/components/schemas/ServerErrorResponse' description: Bad gateway error. '503': content: text/event-stream: schema: $ref: '#/components/schemas/ServerErrorResponse' description: Service unavailable. security: - HTTPBearer: [] components: schemas: GenerateDraftSchema: properties: steer_text: anyOf: - maxLength: 500 type: string - type: 'null' title: Steer Text title: GenerateDraftSchema type: object AiDraftReadSchema: properties: ask_reason: anyOf: - $ref: '#/components/schemas/AskReason' - type: 'null' description: >- Why this reply is waiting for a person instead of going out on its own, for example `low_confidence`, `always_ask`, `unverified_sender`, `different_sender` or `daily_cap`. Null when nothing held it back, and on a reply a person asked for. body_text: title: Body Text type: string confidence: anyOf: - $ref: '#/components/schemas/AiConfidence' - type: 'null' confidence_reason: anyOf: - type: string - type: 'null' title: Confidence Reason confidence_score: anyOf: - type: integer - type: 'null' description: >- INB-16: the model's 0-100 calibration, clamped into the categorical band. Null for drafts that predate the field. Presentation only — every decision reads the categorical confidence. title: Confidence Score generated_body_text: title: Generated Body Text type: string id: format: uuid title: Id type: string outbound_message_id: anyOf: - format: uuid type: string - type: 'null' description: >- The agent's message this reply came from, when a sequence's agent wrote it. Edits, ratings and the send are recorded against that message too, which is how the agent learns from them. Null on a reply written in the default voice for a conversation outside any sequence. title: Outbound Message Id rating: anyOf: - $ref: '#/components/schemas/DraftRating' - type: 'null' description: 'INB-06: the human verdict on this draft, when one was given.' send_after: anyOf: - format: date-time type: string - type: 'null' description: >- When a `scheduled` reply goes out on its own, unless a person discards or edits it first. Kept once the reply is sending or sent. Null on a reply that was never scheduled, and on one taken off the schedule, by a person editing it or by a check that held it back for a person (see `ask_reason`). title: Send After sent_by_kind: anyOf: - $ref: '#/components/schemas/MessageSentByKind' - type: 'null' description: >- Who sent the reply: `person` when someone approved it, `system` when the agent sent it on its own. Null until it is sent. source: $ref: '#/components/schemas/AiDraftSource' status: $ref: '#/components/schemas/AiDraftStatus' description: >- Where this reply stands. `pending_approval`: waiting for a person to approve, edit or discard it. `scheduled`: the agent will send it on its own at `send_after` unless a person discards or edits it first. `sending`: it is leaving now and can no longer be changed. `send_unknown`: it was sent, but the mailbox has not confirmed it yet; it settles on its own as `sent` or `send_failed`. `send_failed`: it did not go; write a fresh reply. `sent`: it went out. `discarded`: a person rejected it or asked for a new version. `stale`: the contact wrote again before anyone answered it. steer_text: anyOf: - type: string - type: 'null' title: Steer Text subject: anyOf: - type: string - type: 'null' title: Subject thread_id: format: uuid title: Thread Id type: string required: - id - thread_id - status - source - steer_text - subject - body_text - generated_body_text - confidence - confidence_reason title: AiDraftReadSchema type: object UnauthorizedResponse: properties: detail: description: The detail of why the request is unauthorized. title: Detail type: string required: - detail title: UnauthorizedResponse type: object ForbiddenResponse: properties: detail: description: The detail of why the request is forbidden. title: Detail type: string required: - detail title: ForbiddenResponse type: object UnprocessableEntityResponse: properties: detail: description: >- The detail of the error. This is the error message to show to the user. You should also use the issues field to show more detailed error messages per field. Use this field for a single liner error message. title: Detail type: string issues: description: >- The issues that caused the error. There can be multiple issues. Each issue is a user-facing error message. items: $ref: '#/components/schemas/UnprocessableEntityIssue' title: Issues type: array required: - detail - issues title: UnprocessableEntityResponse type: object ServerErrorResponse: properties: detail: description: The detail of why the server returned an error. title: Detail type: string required: - detail title: ServerErrorResponse type: object AskReason: description: >- Why a message that could have sent unattended was put in front of a person. One vocabulary for every gate between an agent writing a message and the message leaving: what the agent itself said about the draft (``LOW_CONFIDENCE``, ``FINDINGS``, ``INSUFFICIENT_DATA``, ``POLICY``), what the sender or the cadence asked for (``ALWAYS_ASK``, ``SEQUENCE_MANUAL``, ``STEP_REQUIRES_APPROVAL``), what the pacing rules refused (``DAILY_CAP``, ``CONSECUTIVE_CAP``, ``THREAD_CAP``, ``HOURLY_CAP``, ``ORG_DAILY_CAP``), what the mailbox or the thread made unsafe (``UNVERIFIED_SENDER``, ``DIFFERENT_SENDER``, ``LABEL``, ``STALE_INBOUND``, ``SEQUENCE_GONE``), and what the counterparty turned out to be (``BOT_QUESTION``, ``AUTOMATED_COUNTERPARTY``, ``AUTHENTICATION``, ``LANGUAGE``, ``LINKEDIN_ONLY``). ``KILL_SWITCH`` is the platform switch that stops unattended sending outright, and ``SEND_INTERRUPTED`` an unattended reply whose send never got as far as a provider. Stored on the task so the queue can say why each row is there, and so "why did nothing send" is answerable by counting rows rather than by reading logs. enum: - low_confidence - findings - insufficient_data - policy - always_ask - sequence_manual - step_requires_approval - daily_cap - unverified_sender - suppressed - different_sender - label - consecutive_cap - thread_cap - hourly_cap - org_daily_cap - stale_inbound - sequence_gone - bot_question - automated_counterparty - kill_switch - authentication - language - linkedin_only - send_interrupted title: AskReason type: string AiConfidence: description: >- How sure the model is. Shown on the card; no band sends without a person. enum: - high - medium - low title: AiConfidence type: string DraftRating: description: |- INB-06: the human verdict on a drafted reply. Two values, not five stars — a triaging human will click a thumb, not deliberate over a scale, and "good or bad" is all a feedback loop needs. enum: - good - bad title: DraftRating type: string MessageSentByKind: description: >- Who composed an outbound message — a person, or the product itself. Only meaningful outbound, and NULL on anything inbound. The distinction exists because an agent-written reply has to stay attributable after the fact: "did a human send this" is the first question asked of any send that turns out to have been wrong, and a column is the only place that answer survives. enum: - person - system title: MessageSentByKind type: string AiDraftSource: enum: - auto_drafted - on_demand title: AiDraftSource type: string AiDraftStatus: description: >- Where one drafted reply stands. STALE is its own terminal state rather than a flavour of DISCARDED: "the prospect wrote again before anyone answered" and "a human said no" are different facts, and only the second one says anything about draft quality. The send states past SCHEDULED exist for the same reason SEND_UNKNOWN does: once a draft can leave without a person watching, "we do not know" has to be a state of its own. Collapsing it into SEND_FAILED would have a retry send the same reply twice; collapsing it into SENT would lose a reply silently. enum: - pending_approval - sent - discarded - stale - scheduled - sending - send_unknown - send_failed title: AiDraftStatus type: string UnprocessableEntityIssue: properties: input: anyOf: - type: string - type: integer - type: number - type: boolean - items: {} type: array - additionalProperties: true type: object description: The user's input that caused the error. title: Input loc: description: >- The location of the error. Always contains 1 or 2 elements. The first element is the type of location (e.g. 'body', 'query', 'path', 'header'). The second element is the name of the field that caused the error. If the error is not related to a specific field, the second element is not present. items: type: string title: Loc type: array msg: description: The message of the error. The error message to show to the user. title: Msg type: string type: description: >- The type of the error. Useful for frontend code to handle different types of errors. title: Type type: string required: - loc - msg - type - input title: UnprocessableEntityIssue type: object securitySchemes: HTTPBearer: scheme: bearer type: http ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.