> ## Documentation Index > Fetch the complete documentation index at: https://docs.relace.ai/llms.txt > Use this file to discover all available pages before exploring further. # Compact Trace > Compress an agent trace to only the important details. Relace Compact can process agent traces at **>50k tok/s**, enabling just-in-time compaction without interrupting the user experience. ## Input Formats Send `messages` in any of the three standard API formats — the format is detected automatically: | API Format Docs | Native endpoint | | :------------------------------------------------------------------------------------------------ | :--------------------- | | [OpenAI Chat Completions](https://developers.openai.com/api/reference/chat-completions/overview/) | `/v1/chat/completions` | | [OpenAI Responses](https://developers.openai.com/api/reference/resources/responses/) | `/v1/responses` | | [Anthropic Messages](https://platform.claude.com/docs/en/api/messages) | `/v1/messages` | The compressed `messages` in the response come back in the same format you sent, and can replace your live message list directly. ## OpenAPI ````yaml POST /v1/code/compact openapi: 3.0.1 info: title: Relace API description: API for accessing Relace code generation models. version: 1.0.0 license: name: MIT servers: - url: https://models.relace.ai description: Server for model API endpoints - url: https://api.relace.run description: Server for general infrastructure security: - bearerAuth: [] paths: /v1/code/compact: post: description: Compress an agent trace to only the important details. requestBody: description: Agent trace to compact required: true content: application/json: schema: $ref: '#/components/schemas/CompactRequest' example: messages: - role: user content: >- Find where the rate limiter is configured and raise the ceiling to 200 rps. - role: assistant content: '' tool_calls: - id: call_abc123 type: function function: name: grep arguments: '{"pattern": "rate_limit"}' - role: tool tool_call_id: call_abc123 content: 'config/limits.py:14: RATE_LIMIT_RPS = 100' - role: assistant content: >- Found it — the ceiling is set in config/limits.py line 14. Raising it to 200. target_tokens: 96000 agent_model: gpt-5.5 responses: '200': description: Compressed agent trace content: application/json: schema: $ref: '#/components/schemas/CompactResponse' '400': description: Bad request content: application/json: schema: $ref: '#/components/schemas/Error400' '401': description: Unauthorized content: application/json: schema: $ref: '#/components/schemas/Error401' '404': description: API key not found content: application/json: schema: $ref: '#/components/schemas/Error404' '429': description: Rate limit exceeded headers: X-RateLimit-Limit: schema: type: string description: Rate limit ceiling for the API key X-RateLimit-Remaining: schema: type: string description: Number of requests left for the time window X-RateLimit-Reset: schema: type: string description: Time at which the rate limit resets content: application/json: schema: $ref: '#/components/schemas/Error429' '500': description: Internal server error content: application/json: schema: $ref: '#/components/schemas/Error500' servers: - url: https://models.relace.ai components: schemas: CompactRequest: type: object required: - messages properties: messages: type: array description: >- The agent trace to compact, in OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages format. The format is detected automatically, and the compressed trace is returned in the same format. items: type: object description: A message in the same format as the rest of the trace target_tokens: type: integer description: >- Approximate token budget for the retained context, in your agent model's tokens. Defaults to 96k tokens. Counts are computed with a heuristic lookup table to reduce API latency; your provider's count may differ slightly. agent_model: type: string description: >- The model that generated the trace, as its API model id (e.g. `claude-fable-5`, `gpt-5.5`, `grok-4.5`). Relace uses this to apply model-specific compaction improvements and count tokens more accurately. CompactResponse: type: object properties: messages: type: array description: The compressed agent trace, in the same format as the input. items: type: object description: A message in the same format as the input usage: type: object properties: prompt_tokens: type: integer description: Number of tokens in the input conversation completion_tokens: type: integer description: Size of the retained view in the server's tokenizer total_tokens: type: integer description: Total number of tokens used description: >- Token usage information for the request, counted in the Relace Compact tokenizer. example: messages: - role: user content: >- Find where the rate limiter is configured and raise the ceiling to 200 rps. - role: assistant content: >- Found it — the ceiling is set in config/limits.py line 14. Raising it to 200. usage: prompt_tokens: 135407 completion_tokens: 17323 total_tokens: 152730 Error400: type: object properties: error: type: string description: Error message example: Invalid JSON in request body Error401: type: object properties: error: type: string description: Error message example: Authorized header required Error404: type: object properties: error: type: string description: Error message example: 'Bad Request: Route not found' Error429: type: object properties: error: type: string description: Error message example: Rate limit exceeded Error500: type: object properties: error: type: string description: Error message example: Error fetching from origin server securitySchemes: bearerAuth: type: http scheme: bearer description: Relace API key Authorization header using the Bearer scheme. ````