Gateway

A gateway is the governed front door between your team's AI tools and the model provider. Tools speak their native protocol to the gateway; the gateway applies governance and forwards the request upstream with the caller's own credentials.

The request path

sequenceDiagram
    participant CC as Claude Code / SDK
    participant GW as Unyform Gateway
    participant DB as Org store (blueprints + vectors)
    participant AN as Anthropic API
    CC->>GW: POST /gw/{id}/v1/messages (your API key)
    GW->>DB: load blueprint + retrieve relevant code
    GW->>GW: inject context, run input policies
    GW->>AN: forward with YOUR key (BYOK)
    AN-->>GW: response / SSE stream
    GW-->>CC: passthrough + audit entry

Bring your own key (BYOK)

The gateway never holds your provider credentials. Whatever key your tool sends — x-api-key or Authorization: Bearer — is forwarded verbatim to the provider. Governance happens around your request, not by impersonating you.

Endpoints

Native Anthropic protocol — what Claude Code and the Anthropic SDK use. Streaming (SSE) passes through unchanged.

Claude Code's preflight token count. Context injection runs here too, so the count reflects what the real request will send.

OpenAI-compatible protocol for tools that speak it.

Calling it directly

curl https://gateway.unyform.ai/gw/<gateway-id>/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "claude-sonnet-4-6", "max_tokens": 1024,
       "messages": [{"role": "user", "content": "Refactor the circuit breaker."}]}'

import anthropic

client = anthropic.Anthropic(
    base_url="https://gateway.unyform.ai/gw/<gateway-id>",
)
msg = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Refactor the circuit breaker."}],
)

What governance adds to each request

1. Blueprint injection — your repo's architecture and conventions, prepended as cached system context.
2. Retrieved code context — semantic search over your org's code embeddings finds the most relevant existing functions, so generated code matches what's already there.
3. Policies — input/output checks against your org's rules.
4. Audit — every request logged with what was injected and what happened.

Gateway lifecycle

Create gateways per team or per environment in Dashboard → Gateways. Each gateway pins: attached blueprints, enabled providers (with per-provider passthrough), and policy configuration. Deactivating a gateway immediately 503s its traffic.