> ## Documentation Index > Fetch the complete documentation index at: https://docs.osvi.ai/llms.txt > Use this file to discover all available pages before exploring further. # Export Chat Sessions > Returns chat sessions with their transcripts and tool calls, oldest first, as JSON or as Markdown ready to hand to Claude Code, claude.ai, a Codex agent or an MCP client. **Enabled per account.** Ask Osvi support to turn it on; until then every call returns `403 feature_disabled`. **Limits.** With no dates the export covers the last 24 hours, and one request covers at most 31 days. A window serves at most 1,000 sessions, 25 per page. Each token may make 10 requests a minute and 1,000 a day, and an account may run 2 exports at once. Every 429 carries a `Retry-After` header. Pull your chat conversations, with every tool call the agent made, into Claude Code, claude.ai, a Codex agent or your own MCP client. Ask for JSON to process the data, or `format=markdown` to get one readable timeline per conversation, ready to hand to an LLM. The export is enabled per account. Ask Osvi support to turn it on. Until then, every call returns `403 feature_disabled`. ## Use it with Claude Code ```bash theme={null} curl -s -H "API-Token: $OSVI_API_TOKEN" \ "https://api.osvi.ai/v1/chat/export?format=markdown&per_page=25" \ -o chats.md claude "Read chats.md. Which tool calls failed, and what did the customer see when they did?" ``` With no dates, the export covers the **last 24 hours**. On claude.ai, attach the same file to a chat. ## Reading every page Each response tells you exactly how to ask for the next one: * **JSON:** send the parameters in `meta.next_page` as they are. It is `null` on the last page. * **Markdown:** the `X-Next-Page` header holds the next page's query string, and the last line of the document repeats it. `next_page` fixes the window to the timestamps page 1 used (`meta.range`). A window like "the last 24 hours" would otherwise move between requests and skip conversations. For the same reason, a request for page 2 or later without both `start_date` and `end_date` is rejected with `range_required`. If you filter by `status` or `mode`, `next_page` also carries a `snapshot_token`. It keeps page boundaries stable while conversations close or change hands during your export. ## Limits | Limit | Value | | - | - | | Window when no dates are sent | Last 24 hours | | Longest window per request | 31 days. A bare date covers that whole day in IST | | Conversations per window | 1,000. Past that, narrow the window: `422 records_cap_exceeded` | | Conversations per page | 10 by default, 25 at most | | Requests per token | 10 a minute and 1,000 a day (UTC) | | Exports running at once | 2 per account. More returns `429 export_busy` | | Time to build one page | 5 seconds. Longer returns `503 export_timeout` | Every `429` carries a `Retry-After` header with the number of seconds to wait. Most HTTP clients and agent tools follow it automatically. Each conversation is bounded so one very long chat can't crowd out the rest of the page: * **Transcript:** 500 messages and 30,000 characters * **Tool calls:** 500 events and 30,000 characters, counting tool names and IDs. Arguments over 4,000 characters are left out, outputs are cut at 8,000, and error messages at 2,000 * **Input and state fields:** 30,000 characters each When a bound applies, the matching `*_truncated` flag is `true`, so you never mistake a cut-off record for a complete one. ## What's in a tool call | Field | Meaning | | - | - | | `name`, `args` | The tool the agent called and the arguments it sent | | `status` | `ok`, `error`, or `null` if no result was recorded (for example, the conversation ended mid-call) | | `output` | What the tool returned to the agent | | `error_message` | Why it failed, when `status` is `error`. `error_message_truncated` is `true` if it was cut short | | `duration_ms` | How long the tool took | Knowledge-base lookups and the internal events behind buttons, carousels and date pickers are not listed as tool calls. Interactive prompts and replies show up in the transcript with `type: "interaction"`. ## OpenAPI ````yaml GET /v1/chat/export openapi: 3.1.0 info: title: OSVI AI API description: REST API for the OSVI AI voice and chat platform license: name: MIT version: 1.0.0 servers: - url: https://api.osvi.ai security: - ApiToken: [] paths: /v1/chat/export: get: summary: Export chat sessions description: >- Returns chat sessions with their transcripts and tool calls, oldest first, as JSON or as Markdown ready to hand to Claude Code, claude.ai, a Codex agent or an MCP client. **Enabled per account.** Ask Osvi support to turn it on; until then every call returns `403 feature_disabled`. **Limits.** With no dates the export covers the last 24 hours, and one request covers at most 31 days. A window serves at most 1,000 sessions, 25 per page. Each token may make 10 requests a minute and 1,000 a day, and an account may run 2 exports at once. Every 429 carries a `Retry-After` header. parameters: - name: start_date in: query required: false description: >- Start of the window. A date (`2026-10-09`, the whole day in IST) or an ISO 8601 timestamp. Defaults to 24 hours before `end_date`. Required, with `end_date`, on pages after the first: copy both from `meta.next_page`. schema: type: string example: '2026-10-09' - name: end_date in: query required: false description: >- End of the window, same formats. Defaults to now. The window may span at most 31 days. schema: type: string example: '2026-10-09' - name: page in: query required: false description: >- Page number. Follow `meta.next_page` rather than counting pages yourself. schema: type: integer default: 1 example: 1 - name: per_page in: query required: false description: Sessions per page (max 25). schema: type: integer default: 10 maximum: 25 example: 25 - name: format in: query required: false description: >- `json`, or `markdown` for one readable timeline per session. The `.md` path suffix works too. schema: type: string enum: - json - markdown default: json example: markdown - name: status in: query required: false description: Only sessions with these statuses. Accepts a comma-separated list. schema: type: string example: closed - name: mode in: query required: false description: >- Only sessions in this mode. `support` means an operator has taken over. schema: type: string enum: - ai - support example: ai - name: agent_uuid in: query required: false description: >- Only sessions handled by this chat agent. Returns `404` if the agent is not on your account. schema: type: string example: agent_Xk2pQ9rT - name: customer_id in: query required: false description: Only one customer's sessions. schema: type: string example: '+919812345678' - name: snapshot_token in: query required: false description: >- Keeps page boundaries stable while sessions change status or mode. Comes back in `meta` when `status` or `mode` is filtered, and is already included in `meta.next_page`. schema: type: string responses: '200': description: >- One page of sessions. In Markdown, pagination comes back in the `X-Total-Count`, `X-Current-Page`, `X-Total-Pages`, `X-Per-Page`, `X-Records-Cap`, `X-Capped`, `X-Range-Start`, `X-Range-End`, `X-Next-Page` (a query string) and `X-Snapshot-Token` headers. content: application/json: schema: type: object properties: success: type: boolean example: true data: type: array items: $ref: '#/components/schemas/ChatExportSession' meta: $ref: '#/components/schemas/ChatExportMeta' text/markdown: schema: type: string example: > # Chat export: 2026-10-11 10:30:00 UTC to 2026-10-12 10:30:00 UTC Page 1 of 9, 212 sessions in this window. --- ## Session 8f3c2a1e-5b7d-4c9a-9e21-6d0f4b2a7c13 - Agent: Order Support (agent_Xk2pQ9rT) - Customer: +919812345678 - Status: closed | Mode: ai | State: resolution - Started: 2026-10-09 10:02:11 UTC | Last message: 2026-10-09 10:04:52 UTC - Messages: 4 | Tool calls: 1 **User** (10:02:11) Where is my order ORD-5521? > Tool call: lookup_order (ok, 184 ms, 10:02:13) > args: {"order_id":"ORD-5521"} > output: {"status":"shipped","eta":"2026-10-11"} **Assistant** (10:02:15) It shipped yesterday and should arrive Friday. --- _Next page: request again with end_date=2026-10-12T10%3A30%3A00Z&format=markdown&page=2&per_page=25&start_date=2026-10-11T10%3A30%3A00Z_ '401': description: Invalid or missing API token content: application/json: schema: $ref: '#/components/schemas/ErrorUnauthorized' '403': description: The export is not enabled for this account. content: application/json: examples: feature_disabled: summary: Not enabled value: success: false error: feature_disabled feature: chat_export_api message: >- Feature disabled: Public chat export API (GET /v1/chat/export). '404': description: The `agent_uuid` filter does not match an agent on your account. content: application/json: examples: agent_not_found: summary: Unknown agent value: success: false error: agent_not_found details: null '422': description: >- The request can't be served as asked. `details.message` says what to change. content: application/json: examples: date_range_too_long: summary: Window longer than 31 days value: success: false error: date_range_too_long details: max_days: 31 message: >- A single export can cover at most 31 days. Split the range into smaller windows. range_required: summary: Later page without dates value: success: false error: range_required details: message: >- Pages after the first need the start_date and end_date from meta.next_page, so the window doesn't move between pages. invalid_date_range: summary: Unreadable date, or end before start value: success: false error: invalid_date_range details: message: end_date is before start_date. records_cap_exceeded: summary: Past the 1,000-session cap value: success: false error: records_cap_exceeded details: records_cap: 1000 message: >- This export serves at most 1000 records per request window — narrow the date range or filters to page through the rest. snapshot_expired: summary: Snapshot expired mid-walk value: success: false error: snapshot_expired details: message: >- This export snapshot has expired, was not found, or no longer matches the requested filters — restart pagination from page 1 without snapshot_token. unsupported_format: summary: Unknown format value: success: false error: unsupported_format details: message: format must be json or markdown. '429': description: >- Too many requests. Wait for the number of seconds in the `Retry-After` header, then retry. content: application/json: examples: rate_limited: summary: Per-minute or daily limit value: success: false error: Rate limit exceeded. Please try again later. details: retry_after: 23 export_busy: summary: Two exports already running value: success: false error: export_busy details: retry_after: 5 message: >- Another export for this account is still running. Wait and retry, one request at a time. headers: Retry-After: description: Seconds to wait before retrying. schema: type: integer '503': description: The page took longer than 5 seconds to build. content: application/json: examples: export_timeout: summary: Timed out value: success: false error: export_timeout details: message: >- This page took too long to build. Use a smaller per_page or a shorter date range. components: schemas: ChatExportSession: type: object description: >- One chat session. Large fields are bounded per session; each `*_truncated` flag says when a bound applied. properties: session_id: type: string format: uuid example: 8f3c2a1e-5b7d-4c9a-9e21-6d0f4b2a7c13 customer_id: type: string example: '+919812345678' customer_id_truncated: type: boolean example: false agent: type: object properties: uuid: type: string example: agent_Xk2pQ9rT name: type: string example: Order Support mode: type: string enum: - ai - support description: '`support` while an operator has taken over.' example: ai status: type: string enum: - active - closed - dormant example: closed current_state: type: string nullable: true example: resolution current_state_truncated: type: boolean example: false help_requested: type: boolean example: false created_at: type: string format: date-time example: '2026-10-09T10:02:11Z' last_message_at: type: string format: date-time nullable: true example: '2026-10-09T10:04:52Z' message_count: type: integer example: 4 user_message_count: type: integer example: 2 assistant_message_count: type: integer example: 2 input_fields: type: object nullable: true description: >- `null` when larger than 30,000 characters (see `metadata_truncated`). example: name: Riya state_fields: type: object nullable: true example: order_id: ORD-5521 metadata_truncated: type: boolean example: false transcript: type: array description: >- Messages in order. At most 500 messages and 30,000 characters per session. items: $ref: '#/components/schemas/ChatExportMessage' transcript_truncated: type: boolean example: false tool_calls: type: array description: >- Tool calls in order. At most 500 events and 30,000 characters per session. items: $ref: '#/components/schemas/ChatExportToolCall' tool_calls_truncated: type: boolean example: false ChatExportMeta: type: object properties: current_page: type: integer example: 1 total_pages: type: integer description: Pages you can reach under `records_cap`. example: 9 total_count: type: integer description: Sessions matching the window and filters, before the cap. example: 212 per_page: type: integer example: 25 records_cap: type: integer example: 1000 capped: type: boolean description: >- `true` when the window holds more sessions than the export serves. Narrow the window to read the rest. example: false snapshot_token: type: string nullable: true description: >- Only when `status` or `mode` is filtered. Already included in `next_page`. example: c2e1f0a4b7d94e3a snapshot_restarted: type: boolean description: Only when `status` or `mode` is filtered. example: false range: type: object properties: start: type: string format: date-time example: '2026-10-11T10:30:00Z' end: type: string format: date-time example: '2026-10-12T10:30:00Z' defaulted: type: boolean description: '`true` when a date was missing and the default window was used.' example: true max_days: type: integer example: 31 next_page: type: object nullable: true description: >- Query parameters for the next page, with the window fixed to this page's `range`. Send them as-is. `null` on the last page. example: page: 2 per_page: 25 start_date: '2026-10-11T10:30:00Z' end_date: '2026-10-12T10:30:00Z' ErrorUnauthorized: type: object properties: success: type: boolean example: false error: type: string example: Invalid API token ChatExportMessage: type: object properties: role: type: string example: user content: type: string example: Where is my order ORD-5521? created_at: type: string format: date-time example: '2026-10-09T10:02:11.087Z' type: type: string enum: - message - interaction description: '`interaction` for a button, carousel or date-picker prompt or reply.' example: message ChatExportToolCall: type: object description: One tool call the agent made during the session, paired with its result. properties: request_id: type: string nullable: true description: The model's tool-call id; links the call to its result. example: call_7Hq2 name: type: string nullable: true example: lookup_order called_at: type: string format: date-time description: When the call was made, in UTC with milliseconds. example: '2026-10-09T10:02:13.412Z' args: type: object nullable: true description: >- Arguments the agent sent. `null` when they exceeded 4,000 characters (see `args_truncated`) or when only the result was recorded. example: order_id: ORD-5521 args_truncated: type: boolean example: false status: type: string nullable: true enum: - ok - error - null description: >- `null` when no result was recorded, for example when the session ended mid-call. example: ok output: type: string nullable: true description: What the tool returned to the agent, clipped to 8,000 characters. example: '{"status":"shipped","eta":"2026-10-11"}' output_truncated: type: boolean example: false error_message: type: string nullable: true description: Why the call failed, clipped to 2,000 characters. example: null error_message_truncated: type: boolean example: false duration_ms: type: integer nullable: true example: 184 securitySchemes: ApiToken: type: apiKey in: header name: API-Token description: >- 16-character API token associated with your OSVI account. Find it in your dashboard under Settings → API. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.