What is dsh-set-model?

A DeepSeek Harness (DSH) plugin that lets the agent switch model presets and thinking depth on its own, either through its own tools or through Jev, which judges each turn's task type and switches tiers automatically.

How does Jev switch the tier?

On the first step of a turn, Jev gets the latest user instruction and the previous reply, truncated to 2,000 characters, plus every declared preset's id, name and when text as candidates. It returns the best preset and a confidence, and below 0.7 nothing moves.

How do I turn Jev switching on?

jevEnabled defaults to false and is independent of the two tool switches. Set it in the profile config or the Web settings page, where you can also set the API key, base URL, model, confidence threshold and test the connection.

What tools does the agent get?

set_model and list_models switch provider, model and reasoning effort freely when enabled; switch_preset and list_presets constrain the agent to declared tiers, and a tier may declare only a reasoning effort so switching just changes thinking depth on the same model.

How much does the Jev path cost?

The README estimates about 200 to 300 ms and roughly $0.000015 per Jev decision through the OpenRouter Decisions API, with a 3-second fuse timeout. Any failure only writes a warning and keeps the current tier.

What safety rails does it ship?

Provider whitelisting, a context-window guard that refuses switching when the session exceeds the target window, clamping across seven effort levels, config warnings that do not break loading, and tools registered to the root agent only.

Jev directory

CommunityJev Integrations & Adapters

Dsh Set Model

A DeepSeek Harness plugin that lets the agent switch model presets and thinking depth on its own, either with the set_model and switch_preset tools or through Jev, which reads each turn's instruction and picks a tier. Jev auto-switching is a separate switch, off by default.

See highlights Open github.com

Stars: 3
Language: TypeScript
License: MIT
Reading time: 2 min
Published by: Community
Author: john-walks-slow
Last verified: Oct 2026

john-walks-slow/dsh-set-model

User repository on GitHub

View on GitHub

Through Jev or through its own tool calls, the agent switches model presets and thinking depth on its own. A DeepSeek Harness plugin: set_model/list_models and switch_preset/list_presets behind independent switches, Jev auto-switching, Plan Mode staging, and a Web settings page.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity github deepseek plugin routing

Works forJev for Routing Jev for Realtime control

Used inJev for Software & DevTools

Languages: TypeScript
Capabilities: Open sourceRequires API key

Highlights

Three independent switches: enableModelTools (default true), enablePresetTools (default false) and jevEnabled (default false).
With jevEnabled, Jev reads the turn's latest instruction (up to 2,000 characters) and picks a preset, but only when confidence clears 0.7.
Tools: set_model and list_models for free switching, switch_preset and list_presets for declared tiers; every switch is expected to give a reason.
Jev calls go to a configurable Decisions API endpoint, typically about 200-300 ms and roughly $0.000015 per decision with a 3-second timeout.
The Web settings page (Settings, Dynamic Model) manages presets, the policy prompt, Plan-mode targets and Jev options with a test-connection button.

Quickstart

bash

dsh plugin --profile web add dsh-set-model
dsh plugin --profile web add github:john-walks-slow/dsh-set-model

Watch out

README reviewed 2026-10-17; the primary README is Chinese with an English translation in the repo, and the roughly $0.000015 per-decision cost and 200-300 ms latency are the author's own figures rather than independent measurements.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.

FacebookClinicnotes

X: OpenRouter Decision Model Rankings

X: herdr + jev, an agent router

X: A group chat bot with decision room routing

From the Jev guides

Original write-ups on what this entry is for.

Decision gates in agent frameworks: what each integration ships

Oct 16, 2026

How Pydantic AI, Mastra, TanStack AI, DSPy, Spring AI, and LiteLLM expose Jev: where the typed decision sits in the loop, what each integration returns, and which big frameworks are only adjacent.

Jev Integrations & Adapters#agents#frameworks#integrations

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai