One command. Ten DeepSeek Harness sessions.

dsh-cli lets Claude Code, Codex, OpenCode, Pi or any other agent start DeepSeek Harness sessions in parallel, wait for them, read the results and follow up. Research batches, material gathering, live in the DSH web UI.

View on GitHub

Unofficial community project · MIT licensed · Node 22.19+ · macOS and Linux · v0.1.0

Your agentClaude CodeCodexOpenCodePi

dsho research … --jsonINDEX.md 10 reports · 87 sources

Every whale is one DeepSeek Harness session. The whale is DeepSeek's logo, used here to show which tool does the work.

Agent first, human readable.

Every command is non-interactive, returns at once, prints compact text or --json and uses stable exit codes. The same commands work for you in a terminal.

$ dsho research "AI agent pricing" "MCP servers" +8 --depth standard -C ~/dsh-workspaces/ai-2026

Batch b-mh2k9x started: 10 jobs, 10 in parallel (in the DSH web UI)

$ dsho status last-batch

Batch b-mh2k9x [running] ai-2026 10 running

01AI agent pricingrunning

02MCP serversrunning

03Parallel workflowsrunning

04Open-weight modelsrunning

05Coding harnessesrunning

06Agent Skillsrunning

07Browser agentsrunning

08Cost per taskrunning

09Prompt cachingrunning

10Orchestrationrunning

$ dsho collect last-batch

INDEX written: ~/dsh-workspaces/ai-2026/INDEX.md

Sources: ~/dsh-workspaces/ai-2026/sources.json (87 URLs)

Real commands. The output is shortened and the progress tracks are drawn for this page.

Watch every session live.

With dsh web running, a batch becomes its own workspace in the DSH web UI. Watch the workers, read along, step in. Without the UI, the same jobs run headless.

Delegate the slow parts.

DeepSeek is fast and cheap, which makes it a great worker pool. dsh-cli is the missing remote control.

Parallel research batches: One worker per question. Each writes report.md and sources.json; dsho collect merges them into one INDEX.md.
Material gathering: One worker per kind: screenshots, scroll videos, news articles, stock footage and photos, with a manifest.json that records source and licence for every file.
Live in the DSH web UI: When dsh web is running, every job is a real session there. A batch becomes its own workspace, every session gets a readable title, and you can watch or step in.
Headless fallback: Without the UI, each job runs in its own headless dsh process with a lean profile that reuses your research MCP servers.
Steer like a human would: wait, status, logs -f, result, files, continue in the same session and folder, cancel.
No daemon: State is plain JSON in ~/.dsho/. Detached runner processes keep jobs alive after the calling shell exits.

The commands.

References work with a job id, batch id, unique prefix, name, last or last-batch. Every command takes --json.

Main commands
`dsho research <q>...`	parallel research workers (--depth quick\|standard\|deep)
`dsho materials <brief>`	one worker per kind (--kinds, --count, --url)
`dsho batch -f tasks.txt`	many jobs in parallel, one folder each
`dsho run <task>`	one job in the background (-w waits and prints the answer)
`dsho wait <ref>...`	block until done (--timeout, --any)
`dsho status [ref]`	overview, or details of a job or batch
`dsho continue <job> <text>`	continue the same session
`dsho collect <batch>`	write INDEX.md, merge manifest.json and sources.json
`dsho capture shot\|scroll <url>`	clean screenshots and 1080p scroll videos

Exit codes
`0`	success
`1`	a job failed, timed out or was cancelled
`2`	usage error
`3`	job or batch not found
`4`	wait timeout reached, or the job is not finished yet
`5`	environment problem: dsh missing, UI unreachable, Playwright missing

Install in one line.

dsho setup creates the headless dsh profile orchestra and installs the dsh-orchestration skill for Claude Code, Codex, OpenCode and Pi. dsho doctor checks everything. You need Node.js 22.19 or later andDeepSeek Harness, logged in once.

npm install -g @deepseek-ai/dsh   # DeepSeek Harness, if missing
npm install -g dsh-cli && dsho setup
dsho doctor
dsho ui start                     # optional: watch jobs live
dsho research "AI agent pricing 2026" "State of MCP servers" -C ~/dsh-workspaces/ai-2026

Only want the skill? npx skills add nicremo/dsh-cli works too, but the skill needs the CLI. Then ask your agent in plain language: “Use DeepSeek Harness to research these 10 topics in parallel and give me a summary.”

Questions and answers.

Is dsh-cli an official DeepSeek tool?

No. dsh-cli is an unofficial community project by Fabian Bitzer. It is not affiliated with, endorsed by or sponsored by DeepSeek. "DeepSeek Harness" is a trademark of DeepSeek and is used here only to describe compatibility.

What do I need?

Node.js 22.19 or later, DeepSeek Harness installed with npm install -g @deepseek-ai/dsh and logged in once, on macOS or Linux. For dsho capture you also need Python 3 with Playwright and ffmpeg.

What does it cost?

dsh-cli is free and MIT licensed. The workers run on your own DeepSeek Harness setup, so model usage is billed through whatever account that setup uses.

Which agents can use it?

Any agent that can run shell commands. dsho setup installs the dsh-orchestration Agent Skill for Claude Code, Codex, OpenCode and Pi. Agents without skill support can read the same manual with dsho guide.

Is it safe to let workers run unattended?

dsh workers can run shell commands, and many setups run dsh with full access and no sandbox. Keep tasks scoped to the workspace and never delegate deletes, deploys, git pushes or anything that touches secrets. Check licence and source in manifest.json before you publish material a worker found.