Chat wire examples¶

Topics: chat, requests, workers

These JSON examples cover native chat requests, OpenAI-compatible clients and text workers serving chat jobs. Each example links to the corresponding field definitions in the chat requests reference or chat front reference.

Ids, keys and kudos values use placeholders such as "<job-id>". Limits and token counts use realistic values. Response bodies have the server's complete key set. The rc, code, type, object, state, finish_reason, kind and reason strings, and all booleans, match the running code. Other values retain the runtime JSON types unless replaced by placeholders.

The integration tests post each request body and check the stated status. Worker requests include only keys read by the route. tests/integration/test_chat_wire_examples.py checks the native requester and worker examples. tests/integration/test_chat_front_wire_examples.py checks the front examples against the front and in-process Horde.

One chat job across the native routes, with the worker's three ways of ending it:

sequenceDiagram
    participant R as Requester
    participant H as Horde
    participant W as Worker
    R->>H: POST /api/v2/generate/chat/async
    H-->>R: 202 from /chat/async, request id
    W->>H: POST /api/v2/generate/text/pop
    H-->>W: 200 from /text/pop, chat job
    W->>H: POST /api/v2/generate/text/progress (stream_progress jobs)
    H-->>W: 200 from /text/progress, accepted_seq
    alt Requester cancels
        R->>H: DELETE /api/v2/generate/chat/status/{id}
        H-->>R: 200 from /chat/status
        W->>H: POST /api/v2/generate/text/progress
        H-->>W: 200 from /text/progress, cancelled true
    else Worker replies
        W->>H: POST /api/v2/generate/text/submit (message)
        H-->>W: 200 from /text/submit, reward
    else Worker faults
        W->>H: POST /api/v2/generate/text/submit (state faulted)
        H-->>W: 200 from /text/submit, reward 0
    end

Code map¶

Body	File	Symbol
Native request	`horde/apis/v2/kobold.py`, `horde/classes/kobold/chat_request.py`	`ChatAsyncGenerate`, `parse_chat_request`
Native status	`horde/apis/v2/kobold.py`	`ChatAsyncStatus`
Chat model list entry	`horde/apis/v2/base.py`	`Models.add_chat_offers`
Front request	`horde/oai/schema.py`	`ChatCompletionRequest`
Front completion	`horde/oai/mapping.py`	`build_completion`
Front stream chunks	`horde/oai/sse.py`	`build_completion_events`
Front errors	`horde/oai/app.py`	`ChatFrontError`, `_wait_error`, `_kudos_cap_error`
Front model card	`horde/oai/mapping.py`	`build_model_card`
Pop request	`horde/apis/v2/kobold.py`	`TextJobPop`, `TextJobPop.validate_chat_declaration`
Chat job payload	`horde/classes/kobold/waiting_prompt.py`	`TextWaitingPrompt.get_pop_payload`
Submit	`horde/apis/v2/kobold.py`	`TextJobSubmit.prepare_chat_submission`
Progress post	`horde/apis/v2/kobold.py`	`TextJobProgress`
Worker details	`horde/classes/kobold/worker.py`	`TextWorker.get_chat_details`
Example bodies in tests	`tests/integration/_fake_chat_worker.py`	`build_pop_body`, `build_chat_submit_body`, `build_faulted_chat_submit_body`

Native requester¶

A requester queues a chat request on POST /api/v2/generate/chat/async, polls or cancels it on /api/v2/generate/chat/status/{id}, and finds the models available to it on GET /api/v2/status/models?type=text&chat=true. The routes are listed in Routes.

Request¶

POST /api/v2/generate/chat/async with OpenAI's chat keys and the horde options object. The keys and options are specified in Request body.

{
  "model": "<model-name>",
  "messages": [
    {"role": "system", "content": "You are a terse assistant."},
    {"role": "user", "content": [{"type": "text", "text": "Name one colour."}]}
  ],
  "temperature": 0.5,
  "stop": "\nuser:",
  "max_tokens": 64,
  "horde": {
    "max_context_length": 4096,
    "context_fit": "truncate_oldest",
    "conformance": "strict",
    "slow_workers": true,
    "max_kudos": 1000
  }
}

Request accepted¶

A queued request returns 202 with its id, which the status route takes (Routes).

{"id": "<request-id>", "kudos": "<kudos>"}

Dry run¶

A request with horde.dry_run returns 200 with its quote in kudos and is not queued. resolved holds the request as the horde sized it (Context adjustments).

{
  "resolved": {
    "estimated_prompt_tokens": 19,
    "context_fit": "grow",
    "prompt": "system: You are a terse assistant.\nuser: Name one colour.",
    "params": {"max_length": 64},
    "models": ["<model-name>"],
    "style": null
  },
  "kudos": "<kudos>"
}

Status while waiting¶

GET /api/v2/generate/chat/status/{id} returns 200. Here the request waits for its one eligible worker. The fields are those of the text status route (Routes).

{
  "generations": [],
  "context_adjustments": null,
  "impossible_reason": null,
  "fault_reason": null,
  "fault_rc": null,
  "finished": 0,
  "processing": 0,
  "restarted": 0,
  "waiting": 1,
  "done": false,
  "faulted": false,
  "wait_time": 6,
  "queue_position": 1,
  "kudos": "<kudos>",
  "is_possible": true,
  "eligible_workers": 1,
  "eligible_worker_threads": 1,
  "might_stall": false
}

Status when done¶

The status of a served request returns 200 with the generation's structured reply, its worker's reported token counts and the charge_basis used to price it. The generation fields are listed in Pricing.

{
  "generations": [
    {
      "message": {"role": "assistant", "content": "Blue."},
      "finish_reason": "stop",
      "native_finish_reason": "eos_token",
      "prompt_tokens": 21,
      "completion_tokens": 3,
      "reasoning_tokens": null,
      "charge_basis": "output_estimate",
      "queue_wait_seconds": 0.06,
      "text": "Blue.",
      "seed": 0,
      "gen_metadata": [],
      "worker_id": "<worker-id>",
      "worker_name": "<worker-name>",
      "model": "<model-name>",
      "state": "ok"
    }
  ],
  "context_adjustments": null,
  "impossible_reason": null,
  "fault_reason": null,
  "fault_rc": null,
  "finished": 1,
  "processing": 0,
  "restarted": 0,
  "waiting": 0,
  "done": true,
  "faulted": false,
  "wait_time": 0,
  "queue_position": 0,
  "kudos": "<kudos>",
  "is_possible": true,
  "eligible_workers": 1,
  "eligible_worker_threads": 1,
  "might_stall": false
}

Status with context adjustments¶

The 200 status of a long history sent with max_context_length 2048 under the default truncate_oldest: the context grew to an online worker's largest declared chat context, and the oldest turns were dropped to fit it. The entries are specified in Context adjustments.

{
  "generations": [],
  "context_adjustments": [
    {"from": 2048, "to": 8192, "reason": "grow"},
    {"from": 19853, "to": 7459, "reason": "truncate_oldest", "dropped_messages": 10}
  ],
  "impossible_reason": null,
  "fault_reason": null,
  "fault_rc": null,
  "finished": 0,
  "processing": 0,
  "restarted": 0,
  "waiting": 1,
  "done": false,
  "faulted": false,
  "wait_time": 6,
  "queue_position": 1,
  "kudos": "<kudos>",
  "is_possible": true,
  "eligible_workers": 1,
  "eligible_worker_threads": 1,
  "might_stall": false
}

NoEligibleWorker¶

An anonymous request with an image and max_tokens 1024, above the free output floor, while the only online worker serving the model does not declare vision, returns 403. The refusal is specified in Free output allowance and impossible_reason in Impossible reason.

{
  "message": "No online worker can serve this request: no online worker serving the model declares vision.",
  "rc": "NoEligibleWorker",
  "impossible_reason": {"kind": "declaration_gap", "missing_declarations": ["vision"], "exceeded_limits": []}
}

UnsupportedChatParameter¶

A request with logprobs returns 400, with the key in the message. The keys and values refused this way are listed in Refused keys and values.

{
  "message": "'logprobs' is not supported on chat requests.",
  "rc": "UnsupportedChatParameter"
}

Chat model list¶

GET /api/v2/status/models?type=text&chat=true returns 200 with an array holding one entry per model an online chat worker serves, shown here with one entry. The chat object is specified in Chat model list.

[
  {
    "chat": {
      "workers": 1,
      "max_context_length": 8192,
      "max_length": 1024,
      "features": ["presence_penalty", "prompt_overflow_refused", "seed", "tool_choice_enforced", "vision"],
      "created": 1793463623
    },
    "performance": 0.0,
    "queued": 0.0,
    "jobs": 0.0,
    "eta": 0,
    "type": "text",
    "name": "<model-name>",
    "count": 1
  }
]

OpenAI-compatible front¶

A client of the front sends POST /v1/chat/completions with a bearer key and lists models on GET /v1/models. The request keys are specified in Request keys, and their use is in Send chat completions through the OpenAI-compatible front.

Front request¶

POST /v1/chat/completions with OpenAI's keys. The horde object of the native request is accepted here as well (Sampling defaults and the horde object).

{
  "model": "<model-name>",
  "messages": [
    {"role": "system", "content": "You are a terse assistant."},
    {"role": "user", "content": "Name one colour."}
  ],
  "max_tokens": 64,
  "temperature": 0.5
}

Completion¶

The front returns 200 with the completion once the worker's reply arrives. x_horde holds the native request id, the worker and the charge. The fields are specified in Response shape.

{
  "id": "chatcmpl-<request-id>",
  "object": "chat.completion",
  "created": 1793463887,
  "model": "<model-name>",
  "choices": [
    {
      "index": 0,
      "message": {"role": "assistant", "content": "Blue."},
      "finish_reason": "stop",
      "native_finish_reason": null,
      "logprobs": null
    }
  ],
  "usage": {"prompt_tokens": 21, "completion_tokens": 3, "total_tokens": 24},
  "system_fingerprint": "fp_704a460ec9",
  "x_horde": {
    "request_id": "<request-id>",
    "worker_id": "<worker-id>",
    "worker_name": "<worker-name>",
    "kudos": "<kudos>",
    "charge_basis": "output_estimate",
    "estimated_usage": false,
    "queue_wait_seconds": 0.28
  }
}

Content chunk¶

The same request with stream: true and stream_options.include_usage returns 200 with a stream of chat.completion.chunk events. After the role chunk, this chunk has the reply's text. The order of the chunks is in Response shape.

{
  "id": "chatcmpl-<request-id>",
  "object": "chat.completion.chunk",
  "created": 1793463889,
  "model": "<model-name>",
  "choices": [
    {
      "index": 0,
      "delta": {"content": "Blue."},
      "finish_reason": null,
      "native_finish_reason": null,
      "logprobs": null
    }
  ],
  "usage": null,
  "system_fingerprint": "fp_704a460ec9"
}

Finish chunk¶

The finish chunk of the same 200 stream has an empty delta, the finish_reason and x_horde.

{
  "id": "chatcmpl-<request-id>",
  "object": "chat.completion.chunk",
  "created": 1793463889,
  "model": "<model-name>",
  "choices": [
    {
      "index": 0,
      "delta": {},
      "finish_reason": "stop",
      "native_finish_reason": null,
      "logprobs": null
    }
  ],
  "usage": null,
  "system_fingerprint": "fp_704a460ec9",
  "x_horde": {
    "request_id": "<request-id>",
    "worker_id": "<worker-id>",
    "worker_name": "<worker-name>",
    "kudos": "<kudos>",
    "charge_basis": "output_estimate",
    "estimated_usage": false,
    "queue_wait_seconds": 0.2
  }
}

Usage chunk¶

The last chunk of the same 200 stream, sent for stream_options.include_usage, has empty choices and the usage. data: [DONE] follows it (Response shape).

{
  "id": "chatcmpl-<request-id>",
  "object": "chat.completion.chunk",
  "created": 1793463890,
  "model": "<model-name>",
  "choices": [],
  "usage": {"prompt_tokens": 21, "completion_tokens": 3, "total_tokens": 24},
  "system_fingerprint": "fp_704a460ec9"
}

Tool-call completion¶

A request offering a get_weather tool returns 200 with the worker's call: content is null and finish_reason is tool_calls. The message shape is specified in Response shape, and the round trip is in Tools and tool_choice.

{
  "id": "chatcmpl-<request-id>",
  "object": "chat.completion",
  "created": 1793463891,
  "model": "<model-name>",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": null,
        "tool_calls": [
          {
            "id": "call_weather_1",
            "type": "function",
            "function": {"name": "get_weather", "arguments": "{\"city\":\"Paris\"}"}
          }
        ]
      },
      "finish_reason": "tool_calls",
      "native_finish_reason": null,
      "logprobs": null
    }
  ],
  "usage": {"prompt_tokens": 21, "completion_tokens": 3, "total_tokens": 24},
  "system_fingerprint": "fp_704a460ec9",
  "x_horde": {
    "request_id": "<request-id>",
    "worker_id": "<worker-id>",
    "worker_name": "<worker-name>",
    "kudos": "<kudos>",
    "charge_basis": "output_estimate",
    "estimated_usage": false,
    "queue_wait_seconds": 0.21
  }
}

No worker supports features¶

A request with reasoning_effort while the only online worker serving the model does not declare that feature returns 404, and the front cancels the native request. The response is specified in No eligible worker.

{
  "error": {
    "message": "No worker can serve this request: no online worker serving the model declares reasoning_effort. The request was cancelled.",
    "type": "not_found_error",
    "param": null,
    "code": "no_worker_supports_features",
    "metadata": {"request_id": "<request-id>"}
  }
}

Kudos cap¶

A request whose kudos estimate is above its horde.max_kudos returns 400, and the horde does not create it. The values in metadata are read from the native ChatKudosCapExceeded message, as specified in Kudos cap.

{
  "error": {
    "message": "The request's kudos estimate at current pricing is 215, above its horde.max_kudos of 1. The estimate assumes the full max_tokens is generated.",
    "type": "invalid_request_error",
    "param": "horde.max_kudos",
    "code": "max_kudos_exceeded",
    "metadata": {"rc": "ChatKudosCapExceeded", "kudos_estimate": 215, "max_kudos": 1}
  }
}

Model list¶

GET /v1/models returns 200 with one model card per text model an online chat worker serves, shown here with one card. The card's fields are specified in Model card fields and read as in Choose a model from the model card.

{
  "object": "list",
  "data": [
    {
      "id": "<model-name>",
      "object": "model",
      "created": 1793463881,
      "owned_by": "aihorde",
      "context_length": 8192,
      "architecture": {
        "modality": "text+image->text",
        "input_modalities": ["text", "image"],
        "output_modalities": ["text"]
      },
      "top_provider": {"context_length": 8192, "max_completion_tokens": 1024},
      "supported_parameters": [
        "max_tokens",
        "temperature",
        "top_p",
        "stop",
        "seed",
        "presence_penalty",
        "tools",
        "parallel_tool_calls",
        "tool_choice",
        "response_format"
      ],
      "x_horde": {"workers": 1, "queued": 0.0}
    }
  ]
}

Worker¶

A text worker pops chat jobs on /api/v2/generate/text/pop, posts progress on /api/v2/generate/text/progress and submits on /api/v2/generate/text/submit. The procedure is Serve chat jobs from a text worker.

Pop¶

Pop request¶

POST /api/v2/generate/text/pop with the chat declaration beside the text fields. chat_features lists three features and two guarantees together. The chat fields are specified in Chat declaration on pop and explained in Declare chat on pop.

{
  "name": "<worker-name>",
  "models": ["<model-name>"],
  "bridge_agent": "<bridge-agent>",
  "amount": 1,
  "max_context_length": 4096,
  "max_length": 512,
  "chat_jobs": true,
  "chat_features": [
    "seed",
    "presence_penalty",
    "vision",
    "tool_choice_enforced",
    "prompt_overflow_refused"
  ],
  "chat_limits": {"max_context_length": 8192, "max_length": 1024},
  "model_id": "<model-name>",
  "image_transports": ["url", "base64"]
}

Chat job¶

The pop returns 200 with chat: true and the conversation in payload.messages. The top-level messages field contains horde messages for the worker's owner. The payload is specified in Chat job pop response.

{
  "payload": {
    "messages": [
      {"role": "system", "content": "You are a terse assistant."},
      {"role": "user", "content": [{"type": "text", "text": "Name one colour."}]}
    ],
    "temperature": 0.5,
    "top_p": 1.0,
    "frequency_penalty": 0.0,
    "presence_penalty": 0.0,
    "rep_pen": 1.0,
    "top_k": 0,
    "min_p": 0.0,
    "stop_sequence": ["\nuser:"],
    "max_length": 512,
    "max_context_length": 2048,
    "n": 1
  },
  "id": "<job-id>",
  "ids": ["<job-id>"],
  "messages": [],
  "ttl": 150,
  "extra_source_images": null,
  "skipped": {},
  "softprompt": null,
  "model": "<model-name>",
  "chat": true
}

Chat job with tools¶

A 200 pop of a request that offers a tool, with a history in which the model already called it once and the client answered. The tool keys are specified in Tools and served as in Serve tools and structured output.

{
  "payload": {
    "messages": [
      {"role": "user", "content": "Weather in Oslo, then Bergen?"},
      {
        "role": "assistant",
        "content": null,
        "tool_calls": [
          {
            "id": "call_oslo",
            "type": "function",
            "function": {"name": "lookup_weather", "arguments": "{\"city\":\"Oslo\"}"}
          }
        ]
      },
      {"role": "tool", "content": "Rain, 7C", "tool_call_id": "call_oslo"}
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "lookup_weather",
          "description": "Look up the weather.",
          "parameters": {"type": "object", "properties": {"city": {"type": "string"}}}
        }
      }
    ],
    "tool_choice": "auto",
    "parallel_tool_calls": false,
    "response_format": {"type": "json_object"},
    "temperature": 0.5,
    "top_p": 1.0,
    "frequency_penalty": 0.0,
    "presence_penalty": 0.0,
    "rep_pen": 1.0,
    "top_k": 0,
    "min_p": 0.0,
    "stop_sequence": ["\nuser:"],
    "max_length": 512,
    "max_context_length": 2048,
    "n": 1
  },
  "id": "<job-id>",
  "ids": ["<job-id>"],
  "messages": [],
  "ttl": 150,
  "extra_source_images": null,
  "skipped": {},
  "softprompt": null,
  "model": "<model-name>",
  "chat": true
}

Chat job with an image¶

A 200 pop for a worker that declares vision and only the base64 transport, so the image part arrives as an inline JPEG data URL. Image parts are specified in Images and served as in Serve images.

{
  "payload": {
    "messages": [
      {"role": "system", "content": "Describe images briefly."},
      {
        "role": "user",
        "content": [
          {"type": "text", "text": "What is in these?"},
          {"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,<base64-jpeg>"}}
        ]
      }
    ],
    "temperature": 1.0,
    "top_p": 1.0,
    "frequency_penalty": 0.0,
    "presence_penalty": 0.0,
    "rep_pen": 1.0,
    "top_k": 0,
    "min_p": 0.0,
    "max_length": 64,
    "max_context_length": 2048,
    "n": 1
  },
  "id": "<job-id>",
  "ids": ["<job-id>"],
  "messages": [],
  "ttl": 150,
  "extra_source_images": null,
  "skipped": {},
  "softprompt": null,
  "model": "<model-name>",
  "chat": true
}

Chat job with offloaded messages¶

A 200 pop of an offloaded request for a worker that declares messages_url: the payload has messages_url and does not have messages. The URL and its body are specified in Messages offload and fetched as in Fetch offloaded messages.

{
  "payload": {
    "messages_url": "<signed-url>",
    "temperature": 1.0,
    "top_p": 1.0,
    "frequency_penalty": 0.0,
    "presence_penalty": 0.0,
    "rep_pen": 1.0,
    "top_k": 0,
    "min_p": 0.0,
    "max_length": 64,
    "max_context_length": 2048,
    "n": 1
  },
  "id": "<job-id>",
  "ids": ["<job-id>"],
  "messages": [],
  "ttl": 150,
  "extra_source_images": null,
  "skipped": {},
  "softprompt": null,
  "model": "<model-name>",
  "chat": true
}

No job¶

A pop with no job for the worker returns 200 with id null, an empty ids and no chat key. The pop loop around it is in Recognise a chat job.

{
  "payload": {
    "max_context_length": 2048,
    "max_length": 80,
    "min_p": 0.0,
    "smoothing_factor": 0.0,
    "dynatemp_range": 0.0,
    "dynatemp_exponent": 1.0
  },
  "id": null,
  "ids": [],
  "messages": [],
  "ttl": null,
  "extra_source_images": null,
  "skipped": {},
  "softprompt": null,
  "model": null
}

Submit¶

Reply¶

POST /api/v2/generate/text/submit with a structured reply and the backend's reported counts. The fields are specified in Structured submit and built as in Submit the reply.

{
  "id": "<job-id>",
  "message": {"role": "assistant", "content": "Blue."},
  "finish_reason": "stop",
  "model_id": "<model-name>",
  "state": "ok",
  "prompt_tokens": 21,
  "completion_tokens": 3,
  "native_finish_reason": "eos_token"
}

Tool-call reply¶

A reply that only calls a tool, so content is null and finish_reason is tool_calls, submitted for the job in Chat job with tools. The call's shape is specified in Structured submit.

{
  "id": "<job-id>",
  "message": {
    "role": "assistant",
    "content": null,
    "tool_calls": [
      {
        "id": "call_bergen",
        "type": "function",
        "function": {"name": "lookup_weather", "arguments": "{\"city\":\"Bergen\"}"}
      }
    ]
  },
  "finish_reason": "tool_calls",
  "model_id": "<model-name>",
  "state": "ok"
}

Reply with reasoning¶

A reply whose model reasoned before answering, with the reasoning in message.reasoning_content and its token count in reasoning_tokens. Both are specified in Reasoning.

{
  "id": "<job-id>",
  "message": {
    "role": "assistant",
    "content": "Blue.",
    "reasoning_content": "The user asks for one colour."
  },
  "finish_reason": "stop",
  "model_id": "<model-name>",
  "state": "ok",
  "prompt_tokens": 21,
  "completion_tokens": 40,
  "reasoning_tokens": 37
}

Reply response¶

Each of the three replies above returns 200 with the worker's earned kudos. The reward is priced as in Pricing.

{"reward": "<kudos>"}

Fault¶

A job that the worker cannot serve is submitted with state faulted, neither generation nor message, and here a fault_reason. The values are listed in Worker fault reasons and their use in Fault a job.

{
  "id": "<job-id>",
  "state": "faulted",
  "model_id": "<model-name>",
  "fault_reason": "prompt_overflow"
}

Fault response¶

A fault returns 200 with a reward of 0, and the job returns to the queue for another worker (Fault a job).

{"reward": 0}

Progress¶

Progress post¶

POST /api/v2/generate/text/progress, sent with the worker owner's apikey header, for a job whose payload has stream_progress: true. The fields are specified in Streaming progress and the posting loop is in Streaming progress.

{"id": "<job-id>", "seq": 1, "delta": {"content": "Blue"}}

Progress accepted¶

An accepted post returns 200 with its seq and cancelled false.

{"accepted_seq": 1, "cancelled": false}

Progress after a cancel¶

The first post after the requester cancelled returns 200 with cancelled true. The worker's response to it is in Streaming progress.

{"accepted_seq": 2, "cancelled": true}

Progress out of sequence¶

A post whose seq skips one returns 409 with the next seq the horde accepts. The other refusals of the route are listed in Streaming progress.

{
  "message": "Expected seq 2 for generation <job-id>.",
  "rc": "ChatProgressOutOfSequence",
  "expected_seq": 2
}

Refusals¶

Every refusal has a message for people and an rc for code. The conditions for each return code are in Declare chat on pop and Submit the reply.

ChatDeclarationInvalid¶

A pop whose model_id is not one of its models returns 400, and the pop does not assign a job.

{
  "message": "model_id 'koboldcpp/model_Q4' has to be one of the worker's models.",
  "rc": "ChatDeclarationInvalid"
}

ChatSubmitInvalid¶

A chat submit without finish_reason returns 400. The other malformed submits get the same return code with their own message.

{
  "message": "A chat job's submission needs 'finish_reason'.",
  "rc": "ChatSubmitInvalid"
}

ChatSubmitNonconformant¶

A text reply to a strict request with tool_choice: "required" returns 400. The checks are specified in Reply verification.

{
  "message": "The request's 'tool_choice' is 'required', so the reply needs 'tool_calls'.",
  "rc": "ChatSubmitNonconformant"
}

ChatModelMismatch¶

A submit whose model_id differs from the job's assigned model returns 400.

{
  "message": "'model_id' has to be '<model-name>', the model this job was assigned.",
  "rc": "ChatModelMismatch"
}

Worker details¶

GET /api/v2/workers/name/<worker-name> without an API key returns 200 with the public details of a worker whose last pop declared chat jobs with a tool temperature limit. accepts_chat_jobs, accepts_prompt_jobs and chat are specified in Worker details.

{
  "requests_fulfilled": 120,
  "kudos_rewards": "<kudos>",
  "kudos_details": {"generated": "<kudos>", "uptime": "<kudos>"},
  "performance": "14.2 tokens per second",
  "threads": 1,
  "uptime": 86400,
  "maintenance_mode": false,
  "nsfw": true,
  "trusted": true,
  "flagged": false,
  "uncompleted_jobs": 0,
  "nonconformant_replies": 0,
  "models": ["<model-name>"],
  "team": {"name": null, "id": null},
  "bridge_agent": "<bridge-agent>",
  "accepts_chat_jobs": true,
  "accepts_prompt_jobs": true,
  "chat": {
    "model_id": "<model-name>",
    "features": [
      "presence_penalty",
      "prompt_overflow_refused",
      "seed",
      "tool_choice_enforced",
      "vision"
    ],
    "limits": {"max_context_length": 8192, "max_length": 1024, "tools_max_temperature": 0.7},
    "image_transports": ["base64", "url"]
  },
  "max_length": 512,
  "max_context_length": 4096,
  "type": "text",
  "name": "<worker-name>",
  "id": "<worker-id>",
  "online": true
}

Chat requests reference: every native field and refusal shown on this page.
Chat front reference: the front's request keys, response and stream shapes, errors and model cards.
Send chat completions through the OpenAI-compatible front: the client side of the front's bodies.
Serve chat jobs from a text worker: the pop loop the worker bodies travel through.
Serve your first chat job with koboldcpp: a first end-to-end run with a real backend.