Skip to main content

What to read next

You ran a benchmark on real hardware and read the result. The rest of the documentation, in a useful reading order:

  • Concepts: one page per object, for the moment you need the whole picture of one thing rather than the next step. Benchmark and Job explain what happened during the tutorial; Runtime licenses separates the two artifacts that both get called a license.
  • How-to guides: problem-shaped recipes. Write a job definition is where you go to add a benchmark the deployment does not carry yet; Set up MCP for Claude hands the whole platform to an AI assistant; Manage members and API keys is the one every team needs on day two.
  • CLI reference: every clika-rt command with its arguments and flags. The CLI is how a benchmark run becomes a step in CI, since clika-rt benchmarks watch exits non-zero when a leg fails.
  • ClikaRT: the inference runtime the platform issues credentials for, and the library you link when you ship the model you measured.
  • Modelverse: the CLIKA model library, for packaged models that run on ClikaRT as they are.

Two things worth doing early with a real team:

  1. Name your runs. The results view is built for comparison, and a run named for the question it answers is worth ten named "Untitled Benchmark".
  2. Give integrations their own API key. A key carries a subset of its owner's permissions, so a CI job can dispatch benchmarks without holding an account's full access. Manage members and API keys covers it.