DeepSeek Harness

A modular, plugin-based infrastructure for building and tracing AI agents.

PlayableOpen source

Description

DeepSeek Harness is an open-source framework for building AI agents where every capability—including models, tools, and UI—is a swappable plugin. Built on the Cordis kernel, it features a traceable session log that allows developers to inspect, fork, and replay reasoning paths and multi-step tool calls.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: A significant open-source agentic infrastructure release from a major AI lab, offering deep modularity and traceability.
Attributed to: @deepseek_ai
Built with: Codex V4 GLM-5.3 Kimi K3 V4-Pro Hermes 3
Links: deepseek.com
View tweet
View GitHub repo
Tags: #agent-framework #open-source #modular #developer-tool

Discovered: X
First posted: X · Aug 15, 2026
Listed: Added 54d ago
Build: Strong
Models: Strong

Prompt

You are a senior software engineer tasked with implementing a feature using DeepSeek Harness and Codex Router. Your goal is to write high-quality, production-ready code while efficiently using the available models and routing logic. Always follow the routing graph: - Use V4-Flash for loops, sub-agents, and repeated tasks - Use V4-Pro for complex reasoning and deep work - Use GLM 5.3 and Kimi K3 for intermediate tasks - Cache repeated paths and avoid unnecessary API calls to frontier models Follow the custom agentic workflow saved in the session. Resume from previous state when possible. Output only the final code changes unless asked for explanations.

Source

Explore

Agents / Skills