Fleet Intelligence — Cross-Tenant Model Insights from All Brainstorm Instances

2026-03-24

apisdk-tssdk-pymcp

What We Built

Fleet Intelligence — aggregated learning from all brainstorm instances. Every user's evaluation improves routing for everyone. Query any model to get capability scores, outcome stats, best-for/weak-at signals, common failure modes, and trending data — all anonymized across tenants.

Why It Matters

Individual users see their own usage. Fleet intelligence shows what ALL users collectively know. "Claude Sonnet is 91% on code-generation but weak at large-context search" — that's not one user's opinion, that's 4,721 evaluations from 312 instances.

How It Works

// Model-specific insights
const insights = await client.fleet.insights("anthropic/claude-sonnet-4-5");
// { fleet_stats: { best_for: ["code-generation"], weak_at: ["large-context search"] } }

// Top 10 models overview
const overview = await client.fleet.overview();
// { models: [...], total_models: 42 }

Competitive Edge

No other AI gateway aggregates anonymized intelligence across all customers. Portkey shows your usage. OpenRouter shows pricing. BrainstormRouter shows what actually works — powered by every user's production data.

Lockstep Checklist

[x] API Routes: src/api/routes/fleet.ts — GET with model filter + overview
[x] TS SDK: packages/sdk-ts/src/resources/fleet.ts — insights() + overview()
[x] Python SDK: packages/sdk-py/src/brainstormrouter/resources/fleet.py — sync + async
[x] MCP Schemas: br_fleet_insights tool in src/mcp/server.ts + manifest
[x] Tests: src/api/routes/fleet.test.ts — 9 test cases