What is the dsh-sieve plugin?

A plugin for DeepSeek Harness (DSH) that does context engineering and token optimisation inside the agent loop: it filters tool output, prunes older context and discloses skills progressively to reduce redundant input tokens.

How does dsh-sieve use Jev?

Jev is one of its two judge models, described as hosted and requiring an API key. The README says the plugin never uses the session's model in Jev's place, and with neither judge configured it loads but changes no context.

What is the Laya option?

The other judge is Laya, a local model for macOS on Apple Silicon, which avoids the hosted Jev API key requirement. The README makes clear a judge must be configured for the plugin to change anything.

How much does sieve reduce token usage?

Offline replay measured a 36.28% aggregate reduction in characters across 100 public agent trajectories. Historical live runs showed 21.2% lower main-model input tokens across 80 paired runs, with cache hit rate moving from 93.4% to 92.9%.

Are the live model results current?

No. The README says the live experiment used the now-removed session-model judge path, and the current Jev / Laya configuration has not been re-measured, so treat the 21.2% figure as historical.

How do I install dsh-sieve?

The digest excerpt does not include install or configuration commands. It shows badges for DSH 0.2.1-alpha.1, Cordis 4.0.5-alpha.1 and Node ^22.19 || >=24, so check the repository README for current setup steps.

What testing and licence does dsh-sieve ship with?

GitHub metadata lists MIT, TypeScript 6.0 and 265 passing tests that need no API keys. The README also highlights zero patches to DSH core and replay-safe cold restore of archived originals.

Jev directory

CommunityJev Integrations & Adapters

Dsh Sieve

A context-engineering plugin for DeepSeek Harness that filters tool output, prunes older context and discloses skills progressively, cutting replay payloads by 36.28% in characters, and relies on Jev or Laya as its judge model.

See highlights Open github.com

Stars: 110
Language: TypeScript
License: MIT
Reading time: 2 min
Published by: Community
Author: Sev7eEn7
Last verified: Oct 2026

Sev7eEn7/dsh-sieve

User repository on GitHub

View on GitHub

dsh-sieve: context engineering & token optimization plugin for DeepSeek Harness (DSH) — tool output filtering, context pruning, progressive skill disclosure. 36% smaller payload in offline replay. DSH 上下文管理与 token 优化节省插件。

Forks: 5
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

Tagscommunity github typescript context-engineering dsh

Works forJev for Realtime control Jev for Code

Used inJev for Software & DevTools

Languages: TypeScript
Capabilities: Open source

Highlights

Filters tool output, prunes older context and discloses skills progressively to cut redundant input tokens in AI coding agents.
Measures a 36.28% aggregate payload reduction in characters across 100 public agent trajectories in offline replay.
Historical live runs on deepseek-flash showed 21.2% lower main-model input tokens across 80 paired runs, using a judge path since removed.
Requires a judge model: Jev (hosted, needs an API key) or Laya (local, macOS on Apple Silicon); the session model is never used.
265 tests pass without API keys, DSH core takes zero patches, and rewritten originals are archived for on-demand retrieval.

Watch out

README reviewed 2026-10-17; the excerpt is truncated early in the Why sieve section and shows no install command or configuration example, so setup steps, hosted Jev pricing and interface details are not covered by this entry.

Questions about this entry

Short answers drawn from the linked project or page.

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