Send fewer tokens upstream. Change one line.
Headroom Cloud is a managed, Anthropic-compatible gateway that compresses the context you send upstream — no SDK swap, no prompt rewrite. Point ANTHROPIC_BASE_URL at Headroom, keep your own key, and ship.
$export ANTHROPIC_BASE_URL="https://gateway.headroom.cloud/v1"Authenticate with Authorization: Bearer hr_live_… — Headroom forwards to Anthropic using your own key.
Wire reduction
10–30%
Latency overhead
<15ms
Requests processed
1:1 metered
Uptime target
99.9%
Representative targets, not measured benchmarks.
Transparent interception.
Headroom sits inline between your app and Anthropic. Requests pass through, the context gets compressed, and responses stream straight back — the wire payload shrinks while your code stays exactly the same.
Optimization Rulesets
Adaptive Compression Policies.
Safe
Conservative compression on history only. System and user messages are never touched, the accuracy guard stays on, and the 8 most recent turns are kept verbatim.
Balanced
Compresses eligible history while your system prompt and current user message stay untouched. Accuracy guard on, and the 6 most recent turns are preserved.
Maximum Savings
Stronger history compression for long-running threads. System compression is off by default and requires an explicit opt-in before it engages.
Est. savings ranges are illustrative and vary by workload.
import Anthropic from '@anthropic-ai/sdk'; const client = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY, baseURL: 'https://gateway.headroom.cloud/v1', defaultHeaders: { 'X-Headroom-Policy': 'balanced', },}); async function analyzeCode(source: string) { const res = await client.messages.create({ model: 'claude-sonnet-4-5', max_tokens: 1024, messages: [{ role: 'user', content: source }], }); return res.content;}No rewrites required.
Keep your SDK, your models, and your prompts. Headroom is a drop-in base URL — everything you already built keeps working, only lighter on the wire.
- Uses your own provider key. Headroom forwards to Anthropic with the key you configure — we never mint or hold model credentials on your behalf.
- Streaming (SSE) pass-through. Responses stream straight through, token for token, with tool calls, stop reasons, and usage all preserved.
- Zero prompt retention by default. We record token counts, content hashes, and timing only — raw prompts and completions are never stored.
Simple, scalable pricing.
Start free on the gateway. Upgrade when you need more throughput, more keys, and shared team controls.
Developer
For solo builds and evaluation.
- 1 workspace
- Limited processed tokens
- Safe + Balanced policies
- Community support
Pro
For production apps that need headroom.
- Higher processed-token allowance
- Multiple API keys
- Advanced rate limits
- Usage alerts
- Full analytics dashboard
Team
For teams sharing one gateway.
- Everything in Pro
- Users & workspaces
- Roles & RBAC
- Shared billing & audit log
- Priority support