Quickstart

This guide covers the full path: issue triage → AI dev dispatch → CI → merged PR. You can stop at any step, every feature is independently useful.

We are in closed beta. There is no open registration, so the steps below start once your account is approved. If it is not yet, request access first, we approve in batches.

1. Sign in, then install the GitHub App

Sign in at app.developerz.ai. An approved account gets the GitHub App install link in the dashboard; grant it access to the repositories you want managed.

Installing the App from GitHub before your account is approved does nothing useful, the repositories simply sit unmanaged until the approval lands.

2. Add .dz/maintainer/maintainer.yml

Create .dz/maintainer/maintainer.yml in your repository (the root .maintainer.yml is the legacy path: still read until it is deprecated, and ignored when the .dz/ file exists):

# yaml-language-server: $schema=https://developerz.ai/schemas/maintainer.v1.json
version: 1

Committing the file is the opt-in, so know what those two lines turn on. Every field is optional, but optional means the schema fills it in for you, not that it stays off. A file carrying only version: 1 enables:

A repo with no maintainer.yml runs hands-off instead: no auto-merge, no releases. The difference is deliberate, writing the file is how you say yes. See the maintainer.yml reference for every field.

To start cautious and open up later, commit this instead. It is the no-file posture, stated explicitly, with everything else on:

# yaml-language-server: $schema=https://developerz.ai/schemas/maintainer.v1.json
version: 1
pr:
  auto_merge: false
release:
  manager: none

Editor autocomplete. The yaml-language-server comment wires the JSON Schema so you get validation and hover docs in VS Code, Cursor, and JetBrains.

3. Add your model key

developerz.ai is BYOK, bring your own key. We never proxy or resell tokens.

In the dashboard, go to Keys (/account/keys) and add your inference key:

Your key is encrypted at rest and never logged. See the security doc.

4. Watch the first loop

Open a new issue in your repo. Within seconds the bot will:

1. Read the issue and check for duplicates.
2. Ask for reproduction steps if required fields are missing (configurable).
3. Qualify the issue: bug, feature, question, or dupe.
4. Hand off to your coding agent if handoff.mode is set.

Every comment the bot posts as the GitHub App ends with the bot disclosure footer, appended once by the comment tools, and every comment is written to the audit log. A comment posted on your own user token carries no bot footer, because it is yours (see the security doc).

5. Enroll a box (AI devs + CI runners)

To run AI coding tasks or self-hosted CI, enroll a VPS or cloud VM:

# On your VPS, mint a token in Fleet (/fleet) → Add box, then:
read -rp 'Coding model (e.g. z-ai/glm-5.3): ' DZ_MODEL && \
  export DZ_MODEL_OVERRIDE="${DZ_MODEL:-${DZ_MODEL_OVERRIDE:-}}" && \
  read -rsp 'Enrollment token: ' DZ_ENROLL_TOKEN && DZ_JOIN="$(mktemp)" && \
  curl -sfL -H "X-Dz-Enroll-Token: $DZ_ENROLL_TOKEN" \
    https://api.developerz.ai/join -o "$DZ_JOIN" && bash "$DZ_JOIN"

The first prompt asks which coding model the box drives with your own key; a fresh box that names none is refused before anything is installed or your token is spent.

The token travels in a header rather than a URL, and the script is downloaded to a temp file and then run, not piped into bash, so a truncated download cannot execute half an installer.

The installer prepares the box, enrols it, and installs and starts the dz-runner service, so the box appears in Fleet (/fleet) within one heartbeat. Every guard on that install skips rather than relaxes, so the script's last line says which state it actually left the box in: box <id> is ONLINE, or, when systemd could not start the unit, ENROLLED, BUT NOT BEATING with the reason printed directly above it, see after the join.

If you would rather not paste a secret at all, register the host and let the platform connect over SSH instead, see bring your own boxes.

Requirements

Ubuntu, on x86_64 / aarch64 / arm64. At least 8 GB of RAM (gated at 7600 MiB, since a nominal 8 GB VM reports about 7751 MiB) and 40 GiB free on /, the installer checks both before it downloads anything and refuses below either floor. The coding CLI runs on your box with its checkouts, its code index and a language toolchain beside it; that is what needs the room. No inbound ports are needed.

6. Dispatch your first task

From Claude Code

Create a PAT at API tokens (/account/tokens) with the write:tasks and read:tasks scopes (file the ask, then watch it), then register the developerz.ai MCP server with Claude Code:

read -s DEVELOPERZ_API_KEY; export DEVELOPERZ_API_KEY  # paste the dev_pat_… value
claude mcp add --transport http developerz \
  https://mcp.developerz.ai/v1/mcp \
  --header "Authorization: Bearer $DEVELOPERZ_API_KEY"

claude mcp add stores the header value in plaintext in ~/.claude.json. To keep the token out of any file, commit a project-root .mcp.json instead, which expands ${DEVELOPERZ_API_KEY} from your environment at launch. Both routes are in MCP clients. Then:

Hey Claude, use the developerz.ai ask verb to file "add pagination to the user
listing endpoint" for owner/my-repo.

Triage decides the ask, dispatches the tasks it mints to your AI devs on your own enrolled boxes, and watches the resulting PR and CI before requesting merge.

From Claude Desktop

Claude Desktop takes a stdio command, so bridge through mcp-remote in your claude_desktop_config.json:

{
  "mcpServers": {
    "developerz": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://mcp.developerz.ai/v1/mcp",
        "--header",
        "Authorization: Bearer ${DEVELOPERZ_API_KEY}"
      ],
      "env": { "DEVELOPERZ_API_KEY": "dev_pat_…" }
    }
  }
}

Then ask Claude Desktop to file an ask the same way.

From a phone (Claude mobile)

Claude mobile on iOS or Android adds developerz.ai as a connector via OAuth. the same flow Claude.ai on the web uses, no PAT to paste. See the MCP client onboarding page for the connector URL, the OAuth metadata document, and which clients are substantiated for mobile today.

For other mobile clients: any MCP-capable iOS or Android client that takes a PAT works the same way the desktop clients do. Point it at https://mcp.developerz.ai/v1/mcp with a Authorization: Bearer dev_pat_… header. No iOS/Android-specific config ships in this repo beyond the OAuth recipe for Claude mobile.

7. Enable CI runners

Change one line in any GitHub Actions workflow:

jobs:
  test:
    runs-on: developerz-2vcpu  # was: ubuntu-latest

Push the change. The next workflow run routes to your enrolled box. Logs are stored for 72 hours and downloadable from the dashboard or via the ci log_url MCP verb.

See the CI runners doc for the full security model and label reference.

8. Enable AI code review

Add a review: block to maintainer.yml (or keep your existing .coderabbit.yaml the reviewer reads both, with the native block taking precedence):

review:
  profile: chill
  anti_noise:
    severity_floor: minor
    max_comments: 30

The reviewer triggers on every push to a PR. See the AI review doc.

What happens next

Every action is written to the append-only audit log. Nothing happens silently.