Developer Portal

Build with BrainstormRouter

OpenAI & Anthropic-compatible API, TypeScript & Python SDKs, MCP server, and full machine-readable discovery. One base URL swap — zero code changes.

TypeScript SDK

Drop-in OpenAI replacement. Works with LangChain, Vercel AI, CrewAI.

npm install brainstormrouter

import BrainstormRouter from "brainstormrouter";

const client = new BrainstormRouter({
  apiKey: "brk_..."
});

const res = await client.chat.completions.create({
  model: "auto",
  messages: [{ role: "user", content: "Hello" }],
});

Python SDK

Drop-in OpenAI replacement. Works with LangChain, LlamaIndex, CrewAI.

pip install brainstormrouter

from brainstormrouter import BrainstormRouter

client = BrainstormRouter(api_key="brk_...")

res = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user",
               "content": "Hello"}],
)

REST API OpenAI & Anthropic-compatible

Use any HTTP client. Same /v1/chat/completions contract as OpenAI.

curl -X POST https://api.brainstormrouter.com/v1/chat/completions \
  -H "Authorization: Bearer brk_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "messages": [{"role":"user","content":"Hello"}],
    "stream": true
  }'

MCP Server

Connect from Claude Desktop, Cursor, Windsurf, or any MCP client.

Endpoint:
  POST https://api.brainstormrouter.com/v1/mcp/connect
Transport: Streamable HTTP
Auth: Bearer API key

65 tools available — routing, memory, security, governance, budget, and more.

br_route_completion — Route a chat completion
br_memory_store — Persistent agent memory
br_get_leaderboard — Model performance rankings
br_replay_decisions — Review routing decisions
View all 65 tools →

AI Discovery

Machine-readable manifests so AI agents and coding assistants can find BrainstormRouter automatically.

GET /v1/discovery — Live capability discovery (call first after auth)
llms.txt — Concise AI-readable site map
llms-full.txt — Complete API reference (all routes)
ai-plugin.json — OpenAI plugin manifest
agents.json — Agent capability discovery
openapi.json — OpenAPI 3.1 spec

Integrations

BrainstormRouter is a drop-in swap for any OpenAI or Anthropic-compatible tool.

Authentication

Create an API key from the dashboard or via the API:

# Create a key
curl -X POST https://api.brainstormrouter.com/v1/api-keys \
  -H "Authorization: Bearer brk_existing_key" \
  -H "Content-Type: application/json" \
  -d '{"name": "my-agent", "permissions": ["completions", "memory.read"]}'

# Use it
curl https://api.brainstormrouter.com/v1/models \
  -H "Authorization: Bearer brk_new_key"

BrainstormRouter also supports Supabase JWT (dashboard), Agent JWT (CAF), mTLS (mesh agents), and SCIM bearer tokens (identity providers).

5 Authentication Methods

Method	Header	Use Case
API Key	`Authorization: Bearer br_live_...`	SDKs, scripts, automation
Supabase JWT	`Authorization: Bearer eyJ...`	Dashboard, browser-based UIs
Agent JWT	`Authorization: Bearer <signed JWT>`	CAF agent identity, CSR exchange
mTLS	`X-Amzn-Mtls-Clientcert`	Mesh agents, zero-trust
SCIM	`Authorization: Bearer <SCIM token>`	Identity providers (Okta, Azure AD)

Auto-Routing Variants

Set model to one of these values and the router picks the optimal endpoint:

MODEL VALUE	STRATEGY	USE CASE
`auto`	Thompson Sampling (UCB1 + Gaussian posterior)	General workloads — learns optimal model per task type
`auto:fast`	Lowest p50 latency	Real-time UX, streaming interfaces
`auto:floor`	Cheapest model above quality threshold	Bulk processing, classification, triage
`auto:best`	Highest quality regardless of cost	Critical reasoning, code review, legal

Prometheus test data: Thompson sampling routes simple tasks to DeepSeek at $0.000005/req and complex tasks to Claude Opus at $0.000156/req — a 31x cost difference, chosen automatically.

Intelligence Headers

Every response includes machine-readable headers describing what happened:

HEADER	VALUE
`X-BR-Model-Requested`	What you asked for (`auto`, `gpt-4o`, etc.)
`X-BR-Model-Resolved`	What was actually used
`X-BR-Provider`	Provider that served the request
`X-BR-Route-Strategy`	Routing strategy used
`X-BR-Estimated-Cost`	Pre-execution cost estimate (USD)
`X-BR-Actual-Cost`	Actual cost after execution
`X-BR-Efficiency`	Estimated / actual cost ratio
`X-BR-Cache-Status`	HIT, MISS, or BYPASS
`X-BR-Guardian-Status`	ok, warning, or blocked
`X-BR-Guardian-Overhead-Ms`	Guardian processing time (target <5ms p95)
`X-BR-Request-Id`	Unique request ID for audit trail
`X-BR-Tenant-Id`	Your tenant ID

Error Recovery

Every error response includes a machine-actionable recovery field:

{
  "error": {
    "type": "budget_exceeded",
    "message": "Daily budget limit of $5.00 reached",
    "recovery": {
      "action": "wait_or_upgrade",
      "retry_after": "2026-03-18T00:00:00Z",
      "alternative": "Use auto:floor for lower-cost routing",
      "dashboard_url": "https://brainstormrouter.com/dashboard"
    }
  }
}

Agents can programmatically read the recovery field to decide their next action without human intervention.