Router: Docs

Quickstart

1. Get a key. Sign in and create one on API Keys. It's shown once, so copy it then. Most new accounts get a small free credit; add more on Credits.
2. Point your client at Router. Base URL https://router.enclave.ai/v1, header Authorization: Bearer <key>. Any OpenAI SDK works.
3. Send a request. cyberouter/auto reads the prompt, picks the task, then picks the best model for it. The opencode and Pi tabs give a ready-made config for those coding agents.

curl https://router.enclave.ai/v1/chat/completions \
  -H "Authorization: Bearer $CYBEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cyberouter/auto",
    "messages": [{"role": "user", "content": "Review this diff for injection flaws before we merge."}]
  }'

import os
from openai import OpenAI

client = OpenAI(base_url="https://router.enclave.ai/v1", api_key=os.environ["CYBEROUTER_API_KEY"])
reply = client.chat.completions.create(
    model="cyberouter/auto",
    messages=[{"role": "user", "content": "Review this diff for injection flaws before we merge."}],
)
print(reply.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://router.enclave.ai/v1", apiKey: process.env.CYBEROUTER_API_KEY });
const reply = await client.chat.completions.create({
  model: "cyberouter/auto",
  messages: [{ role: "user", content: "Review this diff for injection flaws before we merge." }],
});
console.log(reply.choices[0].message.content);

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "enclave": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Enclave Router",
      "options": {
        "baseURL": "https://router.enclave.ai/v1",
        "apiKey": "{env:CYBEROUTER_API_KEY}"
      },
      "models": {
        "cyberouter/auto": {
          "name": "Auto (Enclave Router)",
          "tool_call": true,
          "limit": {
            "context": 131072,
            "output": 16384
          }
        },
        "cyberouter/glm-5.3": {
          "name": "GLM 5.3",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/glm-5.3-flash": {
          "name": "GLM 5.3 Flash",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/glm-5.2": {
          "name": "GLM 5.2",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/deepseek-v4-pro": {
          "name": "DeepSeek V4 Pro",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/deepseek-v4.1-flash": {
          "name": "DeepSeek V4.1 Flash",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/deepseek-v4-flash": {
          "name": "DeepSeek V4 Flash",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/qwen3.8-max": {
          "name": "Qwen3.8 Max",
          "tool_call": true,
          "limit": {
            "context": 1010000,
            "output": 16384
          }
        },
        "cyberouter/kimi-k3": {
          "name": "Kimi K3",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/gpt-oss-120b": {
          "name": "GPT-OSS 120B",
          "tool_call": true,
          "limit": {
            "context": 131072,
            "output": 16384
          }
        },
        "cyberouter/minimax-m3": {
          "name": "MiniMax M3",
          "tool_call": true,
          "limit": {
            "context": 524288,
            "output": 16384
          }
        },
        "cyberouter/inkling": {
          "name": "Inkling",
          "tool_call": true,
          "limit": {
            "context": 262144,
            "output": 16384
          }
        },
        "cyberouter/nemotron-ultra": {
          "name": "Nemotron 3 Ultra",
          "tool_call": true,
          "limit": {
            "context": 262144,
            "output": 16384
          }
        },
        "cyberouter/abliterated-large-v2": {
          "name": "Abliterated Large v2 (GLM-5.3, uncensored)",
          "tool_call": true,
          "limit": {
            "context": 1000000,
            "output": 16384
          }
        },
        "cyberouter/abliterated-large": {
          "name": "Abliterated Large (GLM-5.2, uncensored)",
          "tool_call": true,
          "limit": {
            "context": 1000000,
            "output": 16384
          }
        },
        "cyberouter/abliterated-model": {
          "name": "Abliterated Model (uncensored, vision)",
          "tool_call": true,
          "limit": {
            "context": 262144,
            "output": 16384
          }
        }
      }
    }
  }
}

Save as ~/.config/opencode/opencode.json (or opencode.json in your project; merge the provider block if the file exists), then run /models and pick an Enclave Router model.

{
  "providers": {
    "enclave": {
      "baseUrl": "https://router.enclave.ai/v1",
      "api": "openai-completions",
      "apiKey": "CYBEROUTER_API_KEY",
      "models": [
        {
          "id": "cyberouter/auto",
          "name": "Auto (Enclave Router)",
          "contextWindow": 131072,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/glm-5.3",
          "name": "GLM 5.3",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/glm-5.3-flash",
          "name": "GLM 5.3 Flash",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/glm-5.2",
          "name": "GLM 5.2",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/deepseek-v4-pro",
          "name": "DeepSeek V4 Pro",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/deepseek-v4.1-flash",
          "name": "DeepSeek V4.1 Flash",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/deepseek-v4-flash",
          "name": "DeepSeek V4 Flash",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/qwen3.8-max",
          "name": "Qwen3.8 Max",
          "contextWindow": 1010000,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/kimi-k3",
          "name": "Kimi K3",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/gpt-oss-120b",
          "name": "GPT-OSS 120B",
          "contextWindow": 131072,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/minimax-m3",
          "name": "MiniMax M3",
          "contextWindow": 524288,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/inkling",
          "name": "Inkling",
          "contextWindow": 262144,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/nemotron-ultra",
          "name": "Nemotron 3 Ultra",
          "contextWindow": 262144,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/abliterated-large-v2",
          "name": "Abliterated Large v2 (GLM-5.3, uncensored)",
          "contextWindow": 1000000,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/abliterated-large",
          "name": "Abliterated Large (GLM-5.2, uncensored)",
          "contextWindow": 1000000,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/abliterated-model",
          "name": "Abliterated Model (uncensored, vision)",
          "contextWindow": 262144,
          "maxTokens": 16384
        }
      ]
    }
  }
}

Save as ~/.pi/agent/models.json (merge the enclave entry if the file exists), then run pi --provider enclave --model cyberouter/auto or pick a model with /model.

curl https://router.enclave.ai/v1/responses \
  -H "Authorization: Bearer $CYBEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cyberouter/auto",
    "input": "Review this diff for injection flaws before we merge."
  }'

import os
from openai import OpenAI

client = OpenAI(base_url="https://router.enclave.ai/v1", api_key=os.environ["CYBEROUTER_API_KEY"])
response = client.responses.create(
    model="cyberouter/auto",
    input="Review this diff for injection flaws before we merge.",
)
print(response.output_text)

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://router.enclave.ai/v1", apiKey: process.env.CYBEROUTER_API_KEY });
const response = await client.responses.create({
  model: "cyberouter/auto",
  input: "Review this diff for injection flaws before we merge.",
});
console.log(response.output_text);

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "enclave": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Enclave Router",
      "options": {
        "baseURL": "https://router.enclave.ai/v1",
        "apiKey": "{env:CYBEROUTER_API_KEY}"
      },
      "models": {
        "cyberouter/auto": {
          "name": "Auto (Enclave Router)",
          "tool_call": true,
          "limit": {
            "context": 131072,
            "output": 16384
          }
        },
        "cyberouter/glm-5.3": {
          "name": "GLM 5.3",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/glm-5.3-flash": {
          "name": "GLM 5.3 Flash",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/glm-5.2": {
          "name": "GLM 5.2",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/deepseek-v4-pro": {
          "name": "DeepSeek V4 Pro",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/deepseek-v4.1-flash": {
          "name": "DeepSeek V4.1 Flash",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/deepseek-v4-flash": {
          "name": "DeepSeek V4 Flash",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/qwen3.8-max": {
          "name": "Qwen3.8 Max",
          "tool_call": true,
          "limit": {
            "context": 1010000,
            "output": 16384
          }
        },
        "cyberouter/kimi-k3": {
          "name": "Kimi K3",
          "tool_call": true,
          "limit": {
            "context": 1048576,
            "output": 16384
          }
        },
        "cyberouter/gpt-oss-120b": {
          "name": "GPT-OSS 120B",
          "tool_call": true,
          "limit": {
            "context": 131072,
            "output": 16384
          }
        },
        "cyberouter/minimax-m3": {
          "name": "MiniMax M3",
          "tool_call": true,
          "limit": {
            "context": 524288,
            "output": 16384
          }
        },
        "cyberouter/inkling": {
          "name": "Inkling",
          "tool_call": true,
          "limit": {
            "context": 262144,
            "output": 16384
          }
        },
        "cyberouter/nemotron-ultra": {
          "name": "Nemotron 3 Ultra",
          "tool_call": true,
          "limit": {
            "context": 262144,
            "output": 16384
          }
        },
        "cyberouter/abliterated-large-v2": {
          "name": "Abliterated Large v2 (GLM-5.3, uncensored)",
          "tool_call": true,
          "limit": {
            "context": 1000000,
            "output": 16384
          }
        },
        "cyberouter/abliterated-large": {
          "name": "Abliterated Large (GLM-5.2, uncensored)",
          "tool_call": true,
          "limit": {
            "context": 1000000,
            "output": 16384
          }
        },
        "cyberouter/abliterated-model": {
          "name": "Abliterated Model (uncensored, vision)",
          "tool_call": true,
          "limit": {
            "context": 262144,
            "output": 16384
          }
        }
      }
    }
  }
}

Save as ~/.config/opencode/opencode.json (or opencode.json in your project; merge the provider block if the file exists), then run /models and pick an Enclave Router model.

{
  "providers": {
    "enclave": {
      "baseUrl": "https://router.enclave.ai/v1",
      "api": "openai-completions",
      "apiKey": "CYBEROUTER_API_KEY",
      "models": [
        {
          "id": "cyberouter/auto",
          "name": "Auto (Enclave Router)",
          "contextWindow": 131072,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/glm-5.3",
          "name": "GLM 5.3",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/glm-5.3-flash",
          "name": "GLM 5.3 Flash",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/glm-5.2",
          "name": "GLM 5.2",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/deepseek-v4-pro",
          "name": "DeepSeek V4 Pro",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/deepseek-v4.1-flash",
          "name": "DeepSeek V4.1 Flash",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/deepseek-v4-flash",
          "name": "DeepSeek V4 Flash",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/qwen3.8-max",
          "name": "Qwen3.8 Max",
          "contextWindow": 1010000,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/kimi-k3",
          "name": "Kimi K3",
          "contextWindow": 1048576,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/gpt-oss-120b",
          "name": "GPT-OSS 120B",
          "contextWindow": 131072,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/minimax-m3",
          "name": "MiniMax M3",
          "contextWindow": 524288,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/inkling",
          "name": "Inkling",
          "contextWindow": 262144,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/nemotron-ultra",
          "name": "Nemotron 3 Ultra",
          "contextWindow": 262144,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/abliterated-large-v2",
          "name": "Abliterated Large v2 (GLM-5.3, uncensored)",
          "contextWindow": 1000000,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/abliterated-large",
          "name": "Abliterated Large (GLM-5.2, uncensored)",
          "contextWindow": 1000000,
          "maxTokens": 16384
        },
        {
          "id": "cyberouter/abliterated-model",
          "name": "Abliterated Model (uncensored, vision)",
          "contextWindow": 262144,
          "maxTokens": 16384
        }
      ]
    }
  }
}

Save as ~/.pi/agent/models.json (merge the enclave entry if the file exists), then run pi --provider enclave --model cyberouter/auto or pick a model with /model.

OpenAI-compatible: point any SDK at https://router.enclave.ai/v1 and use Chat Completions or the stateless Responses API. cyberouter/auto picks the task and the best model; or pick one from Models.

Choosing a model

Auto. cyberouter/auto classifies the prompt into one of the tasks below, then routes like that task's alias. The receipt says which task it chose and how sure it was. A prompt with no task signal goes to triage.
Task aliases. Name the job and Router picks the model that scores highest on it among endpoints that are up and match your options.
A specific model. Send a model id from Models, like cyberouter/glm-5.3. Router only picks the host. Add "task": "vuln_discovery" to rank hosts by that task's score.

Alias	For
`cyberouter/vuln-discovery`	Find bugs in code, diffs, and configs.	Ranking
`cyberouter/exploit-dev`	Build and debug proofs of concept for authorized testing.	Ranking
`cyberouter/remediation`	Write fixes, patches, and mitigations.	Ranking
`cyberouter/triage`	Sort, summarize, and score findings. Small and fast.	Ranking

Scores and their sources are on Rankings. GET /v1/models lists cyberouter/auto, the task aliases, and every model (add ?task= to filter), and GET /v1/models/{author}/{slug}/endpoints lists a model's hosts.

Provider options

Optional provider object on any request. Defaults are the strict choice. Settings holds account-wide defaults for sort, ZDR, fallbacks and blocked hosts, plus provider locks that pin a model to one host (like a preset). A request's own options win (only or order overrides a lock), except that a host blocked in Settings stays blocked. If a locked host can't serve a request, it fails rather than switching hosts.

Field	Default	What it does
`zdr`	`true`	Only use hosts with zero data retention. If none is up, the request fails instead of falling back to one that keeps data.
`data_collection`	`deny`	Skip hosts that may train on or keep your data.
`sort`	`task_perf`	Rank by task score, or by `price`, `latency` or `throughput`. Task aliases and auto default to task score.
`only` / `ignore`	none	Allow or skip hosts by id: `together`, `fireworks`, `baseten`, `abliteration`.
`order`	none	Preferred host order.
`allow_fallbacks`	`false`	If the chosen host fails, try another host serving the same model, revision and quantization. Router never swaps in a different model.
`max_price`	none	Skip hosts above this USD price per million tokens. A number applies to input and output; `{"prompt": 1, "completion": 4}` sets each.
`preferred_max_latency`	none	Skip hosts whose typical time to first token is above this many milliseconds.
`require_parameters`	`true`	When the request has `tools` or a JSON `response_format`, only use hosts that support them.

{
  "model": "cyberouter/vuln-discovery",
  "messages": [{ "role": "user", "content": "Find the bug in this function." }],
  "provider": { "zdr": true, "only": ["together", "baseten"], "allow_fallbacks": true }
}

Routing receipt

On every response. Successful replies carry provider (who served it) and cyberouter (why). Routing, billing and host errors carry cyberouter too, so you can see what was filtered out.
Key fields. cyberouter.task and task_source (caller, auto or none), chosen (model, host, ZDR), criteria (sort, task score and its source, filters applied), attempts (each host tried and how it went), auto (classifier and confidence) and credits (cost and balance).
Look it up later. Every completion returns an X-Generation-Id header. GET /v1/generation?id= returns the receipt, cost and latency; GET /v1/generations lists recent ones with totals. The same data is on Logs.

Request parameters

Sampling. temperature (0–2), max_tokens / max_completion_tokens, top_p, top_k, min_p, top_a, frequency_penalty and presence_penalty (−2–2), repetition_penalty, seed, stop (up to 4), logit_bias, logprobs, top_logprobs (0–20), reasoning_effort. Out-of-range values get a 400 rather than being clamped.
Tools and JSON. tools, tool_choice and response_format only go to hosts that support them, so they're never silently dropped (unless you set require_parameters: false).
Streaming. "stream": true returns server-sent events. The last chunk carries usage with the cost.
Reasoning. Reasoning models return their thinking in message.reasoning (delta.reasoning when streaming).
Retries. Responses aren't cached and Idempotency-Key isn't supported, so each retry is a new, billed request.

Responses API

POST /v1/responses takes the same models, task and provider as chat, with input, instructions, function tools, text.format and streaming.
Stateless. Router keeps no prompts or outputs, so resend the whole conversation as input each turn. store: true, previous_response_id, conversation, background and stored prompts return 400 stateful_not_supported.

Errors

Errors use the OpenAI shape: {"error": {"message", "type", "code"}}.

Status	Code	Meaning
400	`invalid_request`, `invalid_json`	The body is malformed or a parameter is out of range. The message says which.
400	`unknown_vendor`	A host id in `only`, `ignore` or `order` doesn't exist.
400	`context_length_exceeded`	The prompt plus `max_tokens` doesn't fit the model.
400	`task_conflict`, `auto_unclassified`	The task doesn't match the alias, or auto got an empty prompt.
400	`stateful_not_supported`	A Responses feature that needs stored state.
400	`unsupported_input`, `unsupported_content`, `unsupported_tool`	Responses input Router can't pass on, like `file_id` images or non-function tools.
401	`missing_api_key`, `invalid_api_key`	No key, or a revoked or wrong one.
402	`insufficient_credits`	The balance can't cover the request. Add credits or turn on auto-refill.
402	`budget_exceeded`	Your own spending cap on Credits was hit.
403	`account_blocked`	The account can't use the API.
404	`model_not_found`, `no_matching_endpoint`	Unknown model, or no host is up that meets your options. Router won't quietly loosen them; check `cyberouter.criteria.filters`.
429	`rate_limit_exceeded`	Every host tried was rate limited. Retry after `Retry-After` seconds.
502	`provider_error`	Every host tried failed. `cyberouter.attempts` shows each one.

Credits and billing

Prepaid. 1 credit = $1. Buy on Credits ($5 minimum). Usage is charged at the host's price per token, plus a small markup, and cached prompt tokens cost less where the host discounts them.
What each request cost. usage.cost is what you paid, usage.vendor_cost is the host's price, and cyberouter.credits.balance_usd is what's left.
Holds. Router reserves enough for max_tokens before calling the model and refunds the rest after. Without max_tokens it caps the reply at what your balance can pay for and what fits in the context window.
Controls. On Credits: auto-refill from a saved card, a spending cap that resets monthly or never, and email alerts when spend passes an amount you set or the balance drops below your refill threshold. GET /v1/billing and GET /v1/billing/ledger return the same over the API.

Privacy

No prompts or outputs kept. Router stores routing metadata and token counts, never the content. Requests go only to US-hosted providers, and by default only to hosts that keep no data either.
More in the privacy policy and usage policy.