Harnessarium

See what's behind the harness.

A hands-on lab for looking inside an LLM harness.

A glowing core (the LLM) inside a glass cube (the harness), wired to memory, instructions, tools, skills, MCP, approvals and context; tokens rise from it like bubbles. A glowing core (the LLM) inside a glass cube (the harness), wired to memory, instructions, tools, skills, MCP, approvals and context; tokens rise from it like bubbles.

On its own, a language model only predicts the next tokens: in other words, it answers text. The harness around it connects it to the outside world — memory, instructions, tools, safety — and turns it into an agent. It also decides what the model actually sees.

  • memory
  • instructions
  • tools
  • skills
  • MCP
  • approvals
  • context
  • tokens

Harnessarium will let you inspect the real requests and play with each of these layers yourself.

Coming soon.