Where AgentSight stands today

This is the changelog’s first entry, and it exists to set a baseline rather than announce a release. Everything below is shipped and verified in the product today; anything not listed here is either roadmap or not built.

Instrument. agentsight.init() and two context managers. LLM calls through OpenAI, Anthropic, LangChain and LlamaIndex are captured with no extra code; tools and tasks are decorators. LLM spans carry token counts and model ids only, never prompt or completion text.

Analyse. Five KPI cards and eleven charts on a near real-time dashboard, filterable by date range and environment, plus full transcripts showing the tool and task calls behind each answer. Cost is priced server-side, so history can be restated when a provider changes its rates.

Share. The whole dashboard on your own subdomain under your logo and colours, or embedded inside your product behind a scoped token with origin allow-listing. Field-by-field control over what a client sees; personal data defaults to off in the embed.

Act. Feedback with sentiment and category, and tickets anchored to the exact conversation that went wrong, with priority, tags, environment and a threaded discussion labelled Dev or Reporter.

Still roadmap. AI Agent Analytics — AgentSight’s own agents analysing yours — is not built. It is described on its own page, badged as roadmap, and should not be read as a live capability.

Future entries will be dated releases. If you want them as they land, the feed is at /rss.xml.