Delorun 文档

Install and first run

English

Summary

This page takes you from nothing to your first chat. You run one command, npx delorun. It downloads Delorun, checks that the download is genuine, asks a few questions and sets Delorun up. You then start it with delorun, and it opens in your browser. Everything then runs on your own computer. You need a computer that Delorun supports, Node.js to run the command, and a key for the AI model you want to talk to. A chat needs the key; the studio opens without one, so you can look around first.

Delorun is a community fork of DeepSeek Harness, the open-source program that lets an AI model work on files and run commands on your computer.

Contents

Before you start

  • A supported computer. Linux on x64 or arm64, or Windows through WSL2 (Windows through WSL2). Delorun 0.1.0 does not run on native Windows, and macOS joins in a later release (Roadmap and vision).
  • Node.js 24.9 or newer. Only the install command needs it. Check with node --version.
  • A model key. A key is a secret text that lets Delorun use an AI model on your account. The company or server that runs the model is the model provider. The default is a DeepSeek API key. Other providers work too (Model providers). You can skip the key during setup and add it later.
  • git, which Delorun uses to save your settings before an update (Updates).

Install

Run this in a terminal:

npx delorun

The first run downloads and installs Delorun, then starts setup. Delorun prints the checks described below, then asks the first setup question. Every later npx delorun starts the installed Delorun and downloads nothing.

At the end of setup, Delorun offers Install the delorun command? (runs npm install -g delorun). Say yes, and you can type delorun instead of npx delorun from then on.

Delorun keeps everything in one folder, ~/.delorun, which only you can open (Your Delorun home).

Edge cases
  • If you run macOS, then the launcher stops before it downloads anything and says macOS joins in a later release. Use Linux, or Windows through WSL2.
  • If you run native Windows, then the launcher stops before it downloads anything and says Delorun 0.1.0 runs in WSL2. Follow Windows through WSL2.
  • If your computer runs another system or processor that the release does not list, then the launcher stops with PLATFORM_NOT_RELEASED and installs nothing.
  • If your Node is older than 24.9, then the launcher stops with Delorun needs Node 24.9 or newer to start, and names your version. Install a newer Node from nodejs.org and run npx delorun again.
  • If a check fails, then nothing is installed (What the launcher verifies). Run npx delorun again; if it keeps failing, see Installs and release checks.
  • If you have no network, then the first install cannot download a release. Once a release is installed, starting needs no network.
  • If you set DELORUN_HOME, then the home is created there instead of ~/.delorun (Your Delorun home).
Go further

The launcher takes its own options before any delorun arguments:

OptionWhat it does
--profile <name>Uses another profile than default (Your Delorun home).
--release <version>Installs this release instead of the channel's current one, when no release is installed yet.
--Ends the launcher's options. Everything after it goes to delorun.

Windows through WSL2

Delorun 0.1.0 runs on Windows inside WSL2, the Windows feature that runs Linux next to Windows. It does not run in a native Windows terminal such as PowerShell.

  1. Install WSL2 with its default Ubuntu: open PowerShell and run wsl --install, then restart Windows if it asks.
  2. Open the Ubuntu app from the Start menu. Everything after this happens in that terminal.
  3. Install Node.js 24.9 or newer inside Ubuntu, not in Windows (Before you start).
  4. Run npx delorun there and follow Install.

Keep your projects inside the Ubuntu home folder, such as ~/projects, because files on the Windows drives (/mnt/c) are slow for Linux programs. Open the studio from your Windows browser at the address Delorun prints.

What the launcher verifies

Before it installs anything, the launcher proves the download is the release Delorun's maintainers published. It prints each check:

Resolving the release: the stable channel names 0.1.0.
Checking the signature: 0.1.0 is signed under release root 1.
Checking the release log: verified.
Downloading delorun-0.1.0-linux-x64.tar.gz (…).
Installing release 0.1.0.
  • Signature. Delorun's release keys sign the release. The launcher checks the signatures and that the downloaded file matches. If anything does not match, nothing is installed.
  • Public release log. Every release is also entered in a public, append-only release log. The launcher checks the entry. If the log cannot be reached, the launcher installs the release on its signature alone, prints a warning, and checks the log again at the next start that can reach it.
  • Every start. The launcher compares the installed files with the signed list. A start with a release installed needs no network.

You can run these checks again at any time with delorun verify (Daily use). What you trust when you use Delorun says what you still rely on. The launcher installs from releases.delorun.com, with GitHub Releases and a spare host of Delorun's as fallbacks, and from no other host. It does not check the release's SBOM, and the npm package carries no build attestation.

Edge cases
  • If the log contradicts the release, then the install stops. Do not work around it; see Installs and release checks.
  • If you want a stricter full file check at every start, then set DELORUN_RUNTIME_FULL_CHECK_DAYS to 0. The default hashes every file at least every 7 days; the largest value is 30.

Setup: how you will use Delorun

The first delorun on a new home runs setup. It asks one question at a time, and every question after the first offers ← Back. Nothing is written until you choose Save and set up on the last screen; Cancel there writes nothing.

The first question asks how you will use Delorun. Its answer sets the defaults and decides which questions come next. Each choice shows what it does, who it suits and how to change it later, and the recommended choice is marked:

How will you use Delorun?
● Just me, on this computer (recommended)
○ Also from my phone or another device
○ For scripts and automation

If you are unsure, choose the first. Most people then answer three questions: the model provider, its key, and how much Delorun asks.

You chooseDelorun runsSetup also asks
Just me, on this computerIn the background, and opens in your browser on this computer. Nothing outside this computer can reach it.Nothing more.
Also from my phone or another deviceAs above, plus a private address on your own Tailscale network. A device you approve here with a code can read and reply to chats.Whether to open Delorun from your phone over Tailscale (Reach the studio from your phone).
For scripts and automationWith no studio. You run delorun headless "task" from scripts or CI, with your key in an environment variable such as DEEPSEEK_API_KEY.Neither the import question nor the background question.

Every choice then asks these questions, in this order:

  1. Privacy. A note, no question. It says that Delorun has no telemetry of its own, that DeepSeek Harness's desktop analytics and DeepSeek's session-log upload are off unless you turn them on, and that once a day Delorun checks whether a newer release exists and sends nothing about your installation. The note ends: Delorun checks for new releases once a day. To turn that off before Delorun first starts, choose More settings on the last screen, or run delorun setup privacy later.

  2. Import. Only if you used DeepSeek Harness before and this Delorun home has no chats: Copy your DeepSeek Harness chats, attachments and workspace list? They are only read, and nothing in them changes. The answer starts on No (Switching from DeepSeek Harness). The automation choice never asks it.

  3. Model provider. Which model provider should Delorun use? lists DeepSeek (recommended), More providers (Anthropic, OpenAI, Google Gemini and others, or any server that speaks the OpenAI API) and Choose later.

  4. Model key. The question names the provider's page for creating an API key, then asks How should Delorun get the DeepSeek API key? You can paste it now, and Delorun stores it in your Delorun folder where only you can read it, or use DEEPSEEK_API_KEY, which Delorun reads each time it starts. An environment variable is a named setting that your terminal passes to the programs it starts. If you choose the variable and it is not set in this terminal, setup says so at once and tells you how to set it. Setup checks a typed key with one request to the provider (Model providers).

  5. Prompts. How much should Delorun ask before it changes things? This is the prompts answer. It has three choices:

    ● Ask me first (recommended)
    ○ Ask less (like DeepSeek Harness)
    ○ Never ask (dangerous)

    Ask me first asks before a chat gets full access to a folder, adds a tool server, edits its own instructions or schedules work, and you can allow each once. Ask less (like DeepSeek Harness) behaves as DeepSeek Harness does: few questions, and commands still run in a sandbox (What Delorun protects). Never ask (dangerous) lets any chat switch to full access with no question; it prints a warning, asks you to confirm twice and to type full access everywhere (Consents). It is always listed last.

A review screen then lists every answer, for example for the first choice:

How you use Delorun: just me, on this computer
Release check: on, once a day
Model: DeepSeek, default model. Key: read from DEEPSEEK_API_KEY at each start.
Prompts: Ask me first
Keep running in the background: yes
Remote access: off
Setup name: default, studio port 3080

Nothing is saved until you choose Save and set up.

You should see Save these answers and set up Delorun? with Save and set up, More settings, Cancel and ← Back. Choose Save and set up.

Setup then writes your answers, prints the next steps and starts nothing:

Delorun is set up. Nothing is running yet.

Next:
  1. Set DEEPSEEK_API_KEY in your terminal, then run delorun stop and delorun start if Delorun is running. It is the DeepSeek key you chose.
  2. Start Delorun and open the studio: delorun
     Send a first message in a new chat. The install guide's section on your first chat walks through it: https://docs.delorun.com/guide/install#the-studio-and-your-first-chat

Change an answer later: delorun setup <section> (delorun setup --list shows the sections).
If something looks wrong: delorun doctor.

Step 1 appears only while the key variable you chose is not set. The steps differ by choice. The phone choice adds Add your phone or another device: delorun devices approve, then type the code it shows., or, when Tailscale was not ready, Remote access is off. Fix what the Tailscale note said, then run delorun reach on tailscale. The automation choice replaces step 2 with Run a task from a script or CI: delorun headless "your task" and adds To use the studio as well, start it any time with delorun. When the launcher started setup, the first step installs the delorun command at the same version, with npm install -g delorun@<version>.

Edge cases
  • If you choose Choose later, then no provider is set up and chats cannot answer yet. Setup says so. New chats start on deepseek-flash from DeepSeek, which reads its key from DEEPSEEK_API_KEY. Add a provider later with delorun setup key or in the studio's Settings → Agent → Models.
  • If the provider refuses your key, then setup does not store it unless you choose Continue without a successful check. The review screen then says the key is not verified.
  • If you press Ctrl-C at any question, then setup prints Setup cancelled. Nothing was saved. Run delorun to start again. and exits with code 130. Closing the terminal before you save also writes nothing.
  • If you choose ← Back at the dangerous-mode confirmation, then you return to the prompts question.
  • If you did not choose More settings and port 3080 is already used by another program, then setup does not ask about the port, and the start fails. delorun start prints the home process exited with code 1 before it was running; see its log and the log's path. delorun doctor shows fail port and the fix. Run delorun setup profile to choose another port, then delorun start.
  • If ~/.dsh does not exist, or your Delorun home already has chats, then the import question is skipped.
  • If setup runs without a terminal, then it cannot ask; use delorun setup --yes or --config (see Go further below).
  • If you run setup with --yes and no --provider, then it sets up no model provider and prints how to add one. A script that runs delorun headless needs the key variable set in its own environment, because setup never stores a key it did not ask for.
  • If you run setup as the root user, then Never ask (dangerous) is refused, by the question, by --prompts dangerous and by an answers document.
  • If you chose to run in the terminal (background: No, under More settings), then closing the terminal stops Delorun.
Go further
  • More settings on the review screen asks the rest: whether to check for new releases (Yes by default), whether Delorun keeps running in the background, remote access, and the profile name and the studio's port (3080 by default). Setup says when the port is in use, and asks again. Run setup with another profile name for several setups side by side (Your Delorun home).
  • To change your mind, run delorun setup again. On a finished setup in a terminal it says what it will do, then asks every question again with your current choices selected. Your saved key stays: the key question offers Keep the saved key first and never shows it. The review marks the answers you changed, and only those are written; the rest are reported as kept. Turning dangerous mode on asks both confirmations again. Staying on it asks nothing more. With --yes or --config, a finished setup is left unchanged.
  • Change one answer later with delorun setup <section>. delorun setup --list lists the sections: privacy, import, key, prompts, background, reach and profile.
  • For scripts, delorun setup --yes takes every default and reads the key from DEEPSEEK_API_KEY. delorun setup --config <file> reads the answers from a JSON file, which may name "useCase": "local" | "remote" | "automation". Neither turns on remote access or imports ~/.dsh, and neither starts Delorun: run delorun afterwards. --check-key checks a key that comes from the environment (Model providers).

The studio and your first chat

Run delorun after setup. It starts Delorun and opens the studio in your browser. The studio is Delorun's web page for chatting with the model. It prints the studio address and three commands:

Delorun is running in the background and keeps running after you close this terminal.
Studio: http://127.0.0.1:3080/ (opened in your browser)
  delorun status
  delorun logs --follow
  delorun stop

The studio listens on 127.0.0.1, which only your own computer can reach, unless you turn on remote access. The browser signs in with a link that works once and expires after one minute. delorun open opens the studio again with a fresh link.

The first time, the studio shows a short first-run guide, one step at a time, ending in a chat:

  1. Welcome. Choose the language. Each model provider shows whether its key works. Add or change a provider here.
  2. Look. Choose colours and theme, or the stock look.
  3. Agents. Optionally add Claude Code or Codex as helpers a chat can hand work to.
  4. Recommended plugins. Optionally add plugins, small add-ons that give Delorun more tools.
  5. Phone access. Shown while remote access is not set up. It lists what Tailscale still needs.
  6. Workspace. Choose the workspace for your first chat: the default one, a folder on your computer, or a git address to clone.
  7. Ready. Choose Start a chat.

Every step has Back, Continue and Skip for now. Skip opens no chat. To go through the steps again, choose Settings → General → Getting started → Run setup again. Each step starts on your current choices, such as the mode and theme saved in Settings, and a choice changes only when you change it. On a phone, the Phone access step says that Tailscale is set up on the computer that runs Delorun.

A new chat starts in one of the agent presets, which the studio calls modes: Standard, PTC, Minimal and Creator. In Creator mode, the agent writes plugins and presets with you. Type a message to chat. If you set up no provider or key, the model cannot answer; add one in Settings → Agent → Models.

Edge cases
  • If the browser does not open, then run delorun open, or delorun open --print and open the printed link yourself.
  • If the link says it has expired, then run delorun open again. Each link works once, within one minute.
  • If the browser shows "unauthorized" (HTTP 401) at the studio address, then you opened the address without a sign-in link. Run delorun open.
  • If delorun status says state: not running, then start it with delorun start; if the start fails, read delorun logs and Starting and stopping.
  • If your key is not set, then delorun doctor shows warn key, and a chat reports that the model cannot answer. Fix it with delorun setup key.
Go further

The default workspace

A chat works in a workspace: a folder on your computer whose files the model may read and change. Delorun's default workspace is Delorun/default-workspace in your Documents folder, or in your home folder when your system has no Documents folder. The studio creates it the first time it is used. Add your own project folders in the studio.

Edge cases
  • If none of the candidate folders can be used, for example because one would lie inside your Delorun home, then the start stops with delorun: Delorun found no folder for the default workspace. and lists each folder with the reason. Name a folder in your profile's cordis.patch.yml, as the message shows (Starting and stopping).
  • If you would rather not use the default, then choose another folder in the first-run guide's Workspace step.

A computer without a browser

On a server or over SSH, Delorun cannot open a browser. Forward the studio's port from your own computer, then ask Delorun for a link:

# on your own computer
ssh -L 3080:127.0.0.1:3080 you@server
# on the server, in that SSH session
delorun open --print

delorun open --print prints a link, then This link signs in to the studio once, within one minute; run delorun open --print again for another. Open the link in the browser on your own computer. You should see the studio.

Edge cases
  • If your forward uses another local port, then the link must name it: delorun open --print --port <that port>. A link signs in only at the host and port it names.
  • If the studio uses another port than 3080, then forward that port instead; delorun status prints the address.

Next

For the technically curious

The delorun/docs/contracts/c02-release-manifest.md and the delorun/docs/contracts/c18-release-log.md define every signature and log check the launcher makes. The delorun/docs/contracts/c11-setup-answers.md defines each setup question and what it writes. The delorun/docs/design/first-run.md explains why setup asks what it asks.

本页描述 Delorun 0.0.0。

English

本页目录