clearai-dsh: A Native DSH plugin Bringing the Epistemic Loop to DeepSeek Harness

What Does It Do?

ClearAI is a DSH plugin for DeepSeek Harness that brings the Epistemic Loop to the harness, turning "the model produced an answer" into "the conclusion had to earn its status." It addresses the gap in ordinary agent loops, which track only whether a task is done and never record how a conclusion came to be trusted. The plugin implements framing, hypothesizing, planning, observing, verifying, evaluating, and revising as mechanism rather than advice, deriving state from the session record and giving the model no field in which it could declare a step complete. It is maintained by Clearailhc under the Apache-2.0 license and was last updated in September 2026.

Core Features

• Seven-stage Epistemic Loop: chains question framing, candidate hypotheses, executable plans, observations, evidence verification, evaluation, and revision into one traceable path, compressed at runtime into four beats: plan, execute, observe, reflect.
• Completion computed from evidence: the system decides whether work is finished based on delivered evidence rather than the model's declaration, and above a certain level the doer cannot judge its own result.
• Failures kept as results: a refuted hypothesis is retained as an outcome instead of being deleted, retried, and forgotten like noise.
• Single source of state: all state is derived from the session record rather than stored twice, so progress and phases are computed and cannot drift from what actually happened.
• Preset picker integration: shipped as a native DSH plugin, it exposes a ClearAI preset you can select directly in a session.

How to Use This Plugin?

After enabling it in DSH, open a session and pick ClearAI in the preset picker; the model then works at the rhythm of the Epistemic Loop. Both halves of the plugin are cached inside the running process, so restart the dsh web service afterwards — refreshing the browser is not enough. As you work you will see scope and assumptions, hypotheses, an inspectable plan, and observations with artifacts; to adjust behavior, supply a new frame or constraint in the session. If installation stops because pnpm is missing, follow the official instructions to install pnpm first rather than using corepack.

How to Troubleshoot This Plugin?

If installation stops and reports that pnpm is not on your PATH, DSH manages a profile by driving pnpm, so install pnpm per the official instructions instead of enabling corepack — corepack installs a version router, and the corepack shipped with current Node can fetch a pnpm it is unable to launch. If nothing changes in the interface after enabling, restart the dsh web service as described in the official documentation, since both halves of the plugin are cached inside the running process and a browser refresh will not pick them up.