Call our models —
or serve your agents.

Inference API — models, no agent

Fireworks-style access to our catalog (and your BYO endpoints). Mint a nxi_… key under Dashboard → APIs, then call chat completions.

1. 01Generate an inference key on Dashboard → APIs. Shown once — store it safely.
2. 02Optionally list models to pick a catalog id (or omit model for the default).
3. 03POST to /v1/chat/completions with the key as a bearer token.

POST /v1/chat/completions
Authorization: Bearer nxi_…
Content-Type: application/json

• modelstring

  Catalog model id from GET /v1/models, or the id of one of your BYO custom models. Omit to use the platform default.

• messagesarray

  { role, content } with roles system, user, or assistant. Exactly one of messages / message is required.

• messagestring

  Single user message shortcut (also accepts input).

• streamboolean

  When true, respond with Server-Sent Events.

• temperaturenumber

  Sampling temperature for this call.

• max_tokensnumber

  Cap on completion length for this call.

cURL

curl https://api.nexpher.ai/v1/chat/completions \
  -H "Authorization: Bearer nxi_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepinfra/meta-llama-3.1-8b-instruct",
    "messages": [{"role": "user", "content": "Give me a one-line pitch for my app."}]
  }'

OpenAI Python SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.nexpher.ai/v1",
    api_key="nxi_your_key_here",
)

reply = client.chat.completions.create(
    model="deepinfra/meta-llama-3.1-8b-instruct",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(reply.choices[0].message.content)

Responses follow OpenAI's chat.completion schema — LangChain, Vercel AI SDK, and similar tools work unmodified.

List models

GET /v1/models
Authorization: Bearer nxi_…

Returns the platform catalog plus your registered BYO custom models (owned by the key's account). Use a catalog id or a custom model id as model in completions. BYO runs use your endpoint and do not burn Nexpher inference tokens.

{
  "object": "list",
  "default": "deepinfra/meta-llama-3.1-8b-instruct",
  "data": [
    {
      "id": "deepinfra/meta-llama-3.1-8b-instruct",
      "object": "model",
      "owned_by": "deepinfra",
      "label": "Llama 3.1 8B Instruct"
    }
  ]
}

Agent API — trained assistants

When you need a fixed system prompt, marketplace listing, or fine-tuned behavior, bind a nxa_… key to an agent.

1. 01Create an agent in the builder. Copy its Agent ID from My Agents.
2. 02Generate a per-agent key under Dashboard → APIs.
3. 03Call /v1/agents/:agentId/chat.

POST /v1/agents/:agentId/chat
Authorization: Bearer nxa_…
Content-Type: application/json

• messagestring

  Single user message. Exactly one of message / messages is required (input is accepted as an alias).

• messagesarray

  { role, content } objects. The agent's system prompt is always applied first.

• streamboolean

  When true, respond with Server-Sent Events.

• temperaturenumber

  Defaults to the temperature saved on the agent.

• max_tokensnumber

  Cap on completion length for this call.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.nexpher.ai/v1/agents/<AGENT_ID>",
    api_key="nxa_your_key_here",
)

reply = client.chat.completions.create(
    model="nexpher-agent",          # ignored – the key selects the agent
    messages=[{"role": "user", "content": "Hello!"}],
)
print(reply.choices[0].message.content)

Streaming

Set "stream": true on either endpoint. The response is text/event-stream with OpenAI-style chat.completion.chunk payloads, ending with data: [DONE].

const res = await fetch("https://api.nexpher.ai/v1/chat/completions", {
  method: "POST",
  headers: {
    Authorization: "Bearer nxi_your_key_here",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "deepinfra/meta-llama-3.1-8b-instruct",
    messages: [{ role: "user", content: "Hello" }],
    stream: true,
  }),
});
const reader = res.body.getReader();
const decoder = new TextDecoder();
let buffer = "";
while (true) {
  const { value, done } = await reader.read();
  if (done) break;
  buffer += decoder.decode(value, { stream: true });
  const lines = buffer.split("\n");
  buffer = lines.pop() ?? "";
  for (const line of lines) {
    if (!line.startsWith("data: ") || line === "data: [DONE]") continue;
    const delta = JSON.parse(line.slice(6)).choices[0].delta;
    if (delta.content) appendToUI(delta.content);
  }
}

Errors

Errors return the matching HTTP status with an OpenAI-style body:

{ "error": { "message": "Invalid or disabled API key", "type": "invalid_request_error", "code": 401 } }

Managing keys

Key management uses your logged-in session (Dashboard → APIs). Two key types:

• nxi_… — Inference API (account-scoped, any catalog / BYO model)
• nxa_… — Agent API (bound to one agent)

Inference keys

GET    /inference-api-keys
POST   /inference-api-keys          { "name": "Production" }
#   → returns { …, "apiKey": "nxi_…" } once
GET    /inference-api-keys/:id/usage
PATCH  /inference-api-keys/:id      { "name": "…", "enabled": false }
DELETE /inference-api-keys/:id

Agent keys

GET    /agent-api-keys
POST   /agent-api-keys              { "agentId": "…", "name": "Production" }
#   → returns { …, "apiKey": "nxa_…" } once
GET    /agent-api-keys/:id/usage
PATCH  /agent-api-keys/:id          { "name": "…", "enabled": false }
DELETE /agent-api-keys/:id

• Raw keys are stored only as SHA-256 hashes — we can't recover a lost key, only issue a new one.
• Disabling a key takes effect immediately; requests fail with 401 until re-enabled.
• Inference plan limits: Free 1 · Starter 3 · Pro/Scale unlimited.
• Agent plan limits: Free 1 · Starter 2 · Pro/Scale unlimited (keyed to that agent's billing plan).

Rate limits: each API key allows up to 30 requests per minute by default. Responses include X-RateLimit-Limit and X-RateLimit-Remaining.

Webhooks

Register an HTTPS endpoint and we'll POST signed JSON when things happen. Manage endpoints from Dashboard → APIs → Webhooks.

{
  "id": "evt_9f31c0a2b7d4e5f6a7b8",
  "event": "finetune.succeeded",
  "created": "2026-08-24T09:41:07.000Z",
  "data": {
    "jobId": "…",
    "agentId": "…",
    "displayName": "math-tutor-ft",
    "baseModel": "…",
    "outputModelId": "nexpher/ft/math-tutor"
  }
}

Verifying signatures

Every delivery carries a Nexpher-Signature header of the form t=<unix>,v1=<hex> where hex is HMAC-SHA256 over "<t>.<raw body>" using your endpoint's signing secret.

Quotas & billing

• Inference API calls draw from your personal monthly token allowance — same pool as Chat in the app.
• Agent API calls follow that agent's billing target (you, the team owner, or a shared team plan).
• When the allowance runs out the endpoint answers 402 insufficient_quota.
• BYO / custom models route through your registered endpoint and don't consume Nexpher inference budget.