DeepSeek Harness
A layer on top of other agents, not a replacement: plugins, a log of every step, and Claude Code and Codex called as subagents.
Price and access
- Free?
- The program is free, the model is paid
- Price
- your own key or a DeepSeek account, pay per token
- From Russia
- The program itself has no country limits; you pay the model provider you choose.
- Models
- любой провайдер; свои DeepSeek V4.1 Flash (1017, со зрением) и V4 Pro; вызывает Claude Code и Codex как субагентов
- Licence
- MIT
- Our benchmark
- Models in our benchmark →
What changed in price and access
- DeepSeek API web search and page fetch, including from Harness, Claude Code and Codex, now cost $0.75 per 1,000 successful requests ↗
- DeepSeek-V4.1-Flash-1017 released with vision, tuned for DeepSeek Harness; new Search and Fetch APIs ↗
- DeepSeek API: Chinese national holidays now count as discounted off-peak days, on top of weekends ↗
- DeepSeek V4.1 Flash released: new architecture, 763B parameters, open weights, 1M context, native multimodality; a radically smaller KV cache makes long agent sessions cheaper and faster, and the API is another 50% cheaper off-peak ↗
- DeepSeek V4 Flash Vision, an experimental multimodal version: reads screenshots and diagrams, files up to 64 MB ↗
Changes for this tool in total: 8. The full feed for all agents →
What it gives a manager
- Desktop app for macOS and Windows with Word, spreadsheet and PDF skills: a report or summary gets built without a terminal.
- Scheduled automations (plugin): the agent shows up with its task at the set hour, and every run is listed on its own page.
- The trajectory log shows token spend per step, and a statistics plugin shows model and tool time: you can see where the money went.
- Pick the model per task: routine work on DeepSeek Flash from $0.15 per million input tokens off-peak, hard work on Claude or GPT.
Weak points
- Discussion 461 (Aug 13): an agent in full-access mode wiped a user's entire home directory (Edward Grishin).
- Unvetted plugins: a community audit found only 17 of 122 catalogued plugins install and 10 pass checks; AI RANEZ warns some carry malware.
- V4 Pro cannot see images: for BridgeMind it missed a bug shown in a screenshot that Grok 4.6 fixed in one shot; only Flash has vision.
- Still a preview: as of November 3 every GitHub release is a pre-release; for Nate Herk creator mode kept reverting and context got lost.
How to check this agent’s work
- Compare the trajectory log with the files the agent claims to have read.
- Full-access mode only in a sandbox: in reviews the agent deleted a user's home directory.
- Install only vetted plugins and run the report through a second model – Harness can call Claude Code or Codex itself.
Picking an agent takes an evening. After that the question is not the price but whether you can trust what it did.
You’re on the list
We will write when enrolment opens. Meanwhile, here is what the course covers.
The short course programmeWhat the course covers: programme and price
Strengths
- MIT licence; one npx command opens a browser dashboard on port 3080; any provider, including Bedrock, Vertex, Azure or a custom URL.
- The trajectory log shows every call, the system prompt, reasoning and token spend per step, and a session can be exported for review.
- Nate Herk, same Opus model: a YouTube analytics Excel in about 3 minutes vs 17 in Claude Code, a wiki search in 50 seconds vs 5+ minutes.
- Reads AGENTS.md, CLAUDE.md and Claude skills with no changes; can call Claude Code and Codex as subagents.
Alternatives
Who it is for
- Engineer who already runs several agents and wants a layer over them
- Manager willing to try a preview for a cheap agent that handles documents and spreadsheets
Example from a review: An Excel report on a YouTube channel's analytics
Nate Herk sent the same vague prompt to DeepSeek Harness and Claude Code, both on Opus 5: find the recently pulled channel analytics and build an Excel sheet from them. The Harness came back in about 3 minutes, Claude Code in 17. The Harness version was shorter and easier to read: colour-coded views and engagement, top and bottom 10, monthly trend, tag analysis, though its charts had no labels. Claude Code's was wordier but deeper: all 467 videos, a top 50, more tabs and metrics. He would hand a client the Claude Code file; on a research report using his own skill, the Harness produced 4,400 words with 14 load-bearing sources against Claude Code's 5,000 words and 26.
Reviews (6)
The video authors’ opinions, not our conclusions.
- 100 Hours Testing Deepseek Harness vs. Claude Code. What You Need to Know. · Nate Herk | AI Automation, 23 August 2026The reviewer's view: this is not a free Claude Code but a different harness, faster and endlessly customisable, yet it does not replace Claude Code and he trusts Claude Code's output more.
- Vibe Coding With DeepSeek V4 Pro & DeepSeek Harness. Does It Beat Claude Code? · BridgeMind, 16 August 2026The reviewer spent $5.87 on 114M tokens for the whole video and rates the price 10 of 10, but will not adopt it: the model cannot see the screen and tasks ran 90 minutes.
- DeepSeek Harness – Конец эпохи Claude Code и Codex? · ZProger [ IT ], 18 August 2026The reviewer is enthusiastic about 'everything is a plugin' and the reasoning log; he shows the harness writing a plugin for itself and installing a community terminal UI.
- ПОЛНЫЙ ГАЙД по DEEPSEEK HARNESS: установка и использование | Бесплатная замена Claude Code · Академия нейросетей Сергея Нагорного GORA ACADEMY, 26 August 2026The reviewer would only install it on a disposable 4 GB server, with a second AI dictating commands, and it took about 25 minutes; the program is free, the compute is not.
- Почему все говорят про DeepSeek harness? | Стоит переходить?! · AI RANEZ, 27 August 2026The reviewer will not switch yet, it is too raw; the two real differences from Claude Code and Codex are the trajectory view and customisation, and web search needs its own key.
- DeepSeek Harness, Grok Build, OpenWorker, qm. ОБЗОР 4 HARNESS. ЧТО Я НАШЕЛ В КОДЕ, ВАС УДИВИТ! · Эдвард Гришин | AI и автоматизация бизнеса, 1 September 2026The reviewer advises waiting for a stable release: the code is open but development is closed, issues and pull requests are disabled, and there is no Russian UI and no terminal.
Facts with sources
- Released August 13; within three weeks, about 165 plugins had piled up in the default install, and some of the published ones are malicious. Source (in Russian)
- Deploying it on a clean server took the guide's author about 25 minutes, with a second AI on hand to suggest commands. Source (in Russian)
- Can call Claude Code and Codex as subagents. Source (in Russian)
- Since version 0.2 of October 1 there are desktop apps for macOS and Windows. Source
- DeepSeek-V4.1-Flash-1017 with vision shipped on October 17; since November 1, web search and page fetch via the API cost $0.75 per 1,000 successful requests. Source
- As of November 3 there is no stable release; the 0.2.2-alpha.2 build of October 28 added a Russian interface. Source