Changelog
Changes you can see in the proxy, the dashboard and the docs, newest first.
- Public developer docs at /docs, with a machine-readable OpenAPI description at /openapi.json and an index for AI agents at /llms.txt.
- A security contact at /.well-known/security.txt, and a disclosure policy with a safe harbor for good-faith research at /docs#security.
- Account defaults: set prompt caching once for every agent without its own setting, and for untagged calls.
- Untagged means a call with no X-Vigil-Agent label or an empty one, on every page. An agent you labelled "default" is now an ordinary agent.
- Plus checkout accepts promotion codes.
- Find a call by its x-vigil-request-id on the Calls page: the call, what each optimisation did to it, and its saving — or why Vigil refused it.
- The setup page explains Vigil's request and response headers.
- The Optimise page explains what happens past your plan's limits, and Billing counts optimised agents the way your plan does.
- A kill switch per agent, and an Optimisations panel that lists every optimisation with its status.
- Pricing for GPT-6 Sol and GPT-6 Luna and more OpenAI, Claude and Gemini models; corrected cached-input prices for gpt-4o and gpt-4o-mini.
- The dashboard's server code runs in Frankfurt, beside its database.
- Every response carries x-vigil-request-id; non-streamed responses also carry x-vigil-techniques and x-vigil-saved-usd.
- Per-request controls: X-Vigil-Optimise: off and X-Vigil-Skip.
- No X-Vigil-* request header is forwarded to your provider, X-Vigil-Agent included.
- Every saving on the dashboard is the proxy's own record of each call, made as the call happens.
- In Shadow, OpenAI-compatible providers no longer receive Vigil's cache key.