evaluation infrastructure for ai agents
Build Reliable Agents.
Author evals with Inspect AI, publish the logs with one command, and review every run, with samples, scores, and transcripts in one place.
$ ▍01 · author
Author with Inspect AI
Tasks, datasets, solvers, scorers. Your agent plugs in as a solver, any framework, called over HTTP.
task · dataset · solver · scorer
02 · publish
One command
The evaris CLI uploads the .eval log as an immutable artifact, checksums it, and completes the run.
evaris publish logs --project <id>
03 · review
Every run, rendered
Status, warnings, samples, and scores in the runs UI. Read-only tokens for dashboards and agents.
runs · samples · scores · transcripts
evals are authored with inspect ai, the open-source evaluation framework from the uk ai security institute, mit licensed.
run your first eval in minutes.
pip install inspect-ai · npm install evaris · evaris publish
get started