How it works

One local gateway.
Every agent session,
captured and understood.

Point your agent through Spirehawk with one environment variable. Nothing about how it works changes, but now every session is recorded, costed, and reviewed for you. Your real API keys stay in the vault, added only at call time and never seen by the agent.
YOUR AGENTClaude Code
SPIREHAWKObserve · Measure · Govern
LLMClaude Opus

What happens

From one env line to a full, searchable record of the work.

Route through the local gateway

Point your agent at Spirehawk with a single setting, ANTHROPIC_BASE_URL (or the provider block for Codex). From then on its model calls run through Spirehawk on your own machine. There's nothing to install and no code to rewrite: the agent behaves exactly as it did before, and every reply simply notes that it was routed.

Every session is captured

Prompts, responses, tool steps, token counts, and cost are recorded and encrypted locally, then made searchable and attributed to the ticket behind the work.

Sessions are assessed automatically

A machine engine reads each conversation and labels its state (Running, Needs you, Awaiting instruction, or Completed), so a stalled agent surfaces on its own instead of you watching tabs.

Cost and output on one dashboard

Usage and approximate cost per agent, model, and team, plus the tickets and artifacts each session produced in pilot.

Credentials stay in the vault

Agents carry a Spirehawk capability, never your keys. The proxy swaps the real provider credential in at call time and never forwards the capability upstream.

Three capabilities, one gateway.

Multiply what your agents deliver

Put Spirehawk in the path of your agents today — and turn what your team is already spending into productivity you can see, prove, and grow