Skip to main content
deeda Think is a desktop knowledge/agent platform. Its agent harness drives foundational models from Anthropic, OpenAI, Google Gemini, and local runtimes to the full extent of their capabilities — all governed from a single declarative control plane: workflow markdown, validated and dispatched by Cadence.

Harness Overview

How the pieces fit: Cadence, workflow markdown, runtimes, broker, compiler.

Quickstart

Author and dispatch your first cross-provider workflow.

Runtimes

All 10 runtimes, when to use each, and their capability flags.

Capability Catalog

1,328 of 2,058 capability rows usable today — zero deferred.

​
Design principles

The harness is built on a small set of non-negotiables: one source of truth per concept, no speculative abstractions, behavior-preserving changes by default, and proof over assertion — a capability counts as available only when a real probe exercises it and the evidence is checked in. The docs you are reading inherit that discipline: the data-heavy pages are generated directly from the same catalogs the harness executes against, so they cannot drift from reality.