Skip to main content
Screenshot: The .ade/ directory expanded in a file tree showing ade.yaml, local.yaml, local.secret.yaml, and subdirectories.

​
The .ade/ Directory

When you open a repository in ADE for the first time, ADE initializes a .ade/ directory at the project root. This directory is the single source of truth for all ADE configuration, runtime artifacts, and worktree management for that project. 
.ade/
  ade.yaml              # Shared team config — commit this to git
  local.yaml            # Personal preferences — gitignored
  local.secret.yaml     # API keys and tokens — ALWAYS gitignored
  worktrees/            # Git worktrees for each lane (auto-managed by ADE)
  artifacts/
    packs/              # Context pack snapshots
    checkpoints/        # Session boundary snapshots
    sessions/           # Agent session transcripts
    audit.log           # Append-only audit trail
Never commit local.yaml or local.secret.yaml to git. ADE automatically adds both to .gitignore on initialization, but verify this if you manage .gitignore manually. Committing secrets is irreversible once pushed to a remote.
The worktrees/ and artifacts/ subdirectories are fully managed by ADE. Do not edit files inside them directly — use ADE’s UI or MCP tools instead.

​
Configuration Layering

ADE resolves configuration through three layers applied in order. Lower layers override higher ones.

ade.yaml

Shared team config. Committed to git. Contains project defaults, AI model settings, lane profiles, automation definitions, and trust rules. Every team member reads this file.

local.yaml

Personal preferences. Gitignored. Overrides any value in ade.yaml for your machine only. Contains preferred model, theme, editor settings, and notification preferences.

local.secret.yaml

Secrets. Always gitignored. Contains API keys, OAuth tokens, and MCP server credentials. Only the main process reads this file — the renderer UI never has access to its contents.
When ADE resolves a config value, it checks local.secret.yaml first, then local.yaml, then ade.yaml. If a value exists in local.yaml, it shadows the same key in ade.yaml for that machine.
The layering order means a developer can override the team’s default AI model (set in ade.yaml) by specifying ai.preferredModel in their own local.yaml — without touching the shared config.

​
The Trust Model (SHA-Based Approval)

ade.yaml can contain command definitions (process startup commands, automation executors, test suite runners). Because this file is committed to git, a malicious change in a PR could introduce arbitrary commands that run on your machine. ADE prevents this with a SHA-based approval gate.
1

ADE detects a change

Whenever ade.yaml changes — from a git pull, a teammate’s commit landing via fetch, or a direct edit — ADE immediately detects the difference.
2

ADE shows a diff and pauses

ADE presents a structured diff of the changed fields and suspends execution of any commands defined in the new version. No new processes start, no automations trigger, no agents run commands from the updated config.
3

You review and approve

Review the diff in the approval dialog (Settings → Config → Pending Approval, or the banner that appears automatically). If the changes are expected and safe, click Approve.
4

ADE records the approved SHA

The SHA of the approved ade.yaml is written to your local.yaml. Commands from this version of the config are now permitted to execute.
Screenshot: The config approval dialog showing a diff view of ade.yaml changes with Approve and Reject buttons.
The approved SHA is machine-specific. If two developers need to approve the same change, each must approve it on their own machine. This is intentional — one person’s trust cannot be delegated to another.
To re-approve the currently loaded config (e.g., after a manual edit): Settings → Config → Re-approve current config.

​
ade.yaml — Full Schema Reference

This file is shared with your team. Commit it. Every field is optional except name. 
# .ade/ade.yaml

name: "My Project"
description: "Optional project description shown in ADE's title bar"

# ── AI Defaults ───────────────────────────────────────────────────────────────
ai:
  defaultModel: "claude-opus-4-6"       # or "claude-sonnet-4-6", "gpt-4o", etc.
  defaultBudgetPerSession: 1.00         # USD cap per chat session
  defaultBudgetPerMission: 5.00         # USD cap per mission run
  defaultBudgetPerAutomation: 0.50      # USD cap per automation execution
  monthlyBudgetCap: 100.00              # Hard monthly cap across all agents

# ── Process Definitions ───────────────────────────────────────────────────────
processes:
  - name: "Frontend"
    command: "npm run dev"
    cwd: "."
    port: 3000
    group: "App"
  - name: "API"
    command: "npm run api"
    cwd: "."
    port: 4000
    group: "App"

# ── Test Suites ───────────────────────────────────────────────────────────────
testSuites:
  - name: "Unit Tests"
    run: "npm test"
    watch: "npm test -- --watch"
    ci: "npm test -- --ci --coverage"
  - name: "E2E Tests"
    run: "npm run test:e2e"
    ci: "npm run test:e2e -- --headless"

# ── Lane Profiles ─────────────────────────────────────────────────────────────
lanes:
  profiles:
    - name: "feature"
      basePort: 3001
      env:
        NODE_ENV: "development"
      startCommands: ["npm install"]
    - name: "hotfix"
      basePort: 4001
      env:
        NODE_ENV: "production"

# ── Automation Definitions ────────────────────────────────────────────────────
automations:
  - name: "PR Reviewer"
    trigger:
      type: "git.pr_opened"
    executor: "automation-bot"
    prompt: "Review this PR for code quality, security issues, and adherence to project conventions."
    toolPalette: "read-only"
    budgetCap: 0.50

# ── Trust (managed automatically — do not edit manually) ─────────────────────
trust:
  approvedSha: "abc123..."
All ai.* values are defaults. They can be overridden per-mission, per-worker, or per-automation at creation time. The monthlyBudgetCap is a hard ceiling — once reached, all agent activity pauses until the next billing period or you manually raise the cap in Settings → Budget.
Process definitions describe how to start development servers. ADE uses these to populate the lane startup commands and to generate the process panel in the Lanes view. The group field organizes processes visually in the UI.
ADE uses the run command in agent contexts when a worker needs to execute tests. The watch command is used in lane terminal sessions for continuous testing. The ci command is used during Mission validation phases and automation triggers.
Lane profiles are templates applied when creating new lanes. A profile sets the base port offset, environment variables, and startup commands. When you create a lane and select a profile, all profile values are applied as the lane’s initial configuration.
Automation definitions are shared team-wide. Each automation specifies a trigger event, the executor agent identity, a prompt, the tool palette the agent is allowed to use, and a per-run budget cap. See Automations for the full automation reference.
ADE manages the trust.approvedSha field automatically as part of the approval workflow. Editing it manually will cause ADE to treat the config as unapproved and pause all commands until you complete the approval flow again.

​
local.yaml — Full Schema Reference

This file is yours alone. It is gitignored and never shared. Use it to override team defaults for your own machine. 
# .ade/local.yaml

ai:
  preferredModel: "claude-sonnet-4-6"   # Override the team's default model

editor:
  theme: "dark"                          # dark | light | monokai | solarized-dark | solarized-light | one-dark
  fontSize: 14
  tabSize: 2

terminal:
  shell: "/bin/zsh"
  fontFamily: "JetBrains Mono"
  fontSize: 13

notifications:
  missionComplete: true
  conflictDetected: true
  budgetWarning: true
  prStatusChange: false

# Approval state — managed automatically
trust:
  approvedSha: "abc123..."              # Your machine's approved ade.yaml SHA

​
local.secret.yaml — Full Schema Reference

This file holds your API keys and MCP credentials. It is always gitignored. Only the Electron main process reads it — the renderer (React UI) never has access to its raw contents. 
# .ade/local.secret.yaml

anthropic:
  apiKey: "sk-ant-..."

openai:
  apiKey: "sk-..."

openrouter:
  apiKey: "sk-or-..."

github:
  token: "ghp_..."

linear:
  apiKey: "lin_api_..."

# Optional: local model endpoint (no key needed for Ollama)
localModels:
  baseUrl: "http://localhost:11434/v1"

# External MCP server connections
externalMcp:
  - name: "github"
    transport: "stdio"
    command: "npx"
    args: ["-y", "@modelcontextprotocol/server-github"]
    autoStart: true
    env:
      GITHUB_TOKEN: "ghp_..."

  - name: "ghost-os"
    transport: "http"
    url: "https://ghost-os.local/mcp"
    healthCheckInterval: "30s"
    autoStart: false
local.secret.yaml must never be committed to git. ADE adds it to .gitignore automatically, but also add it manually if your project has nested .gitignore files. Run git check-ignore -v .ade/local.secret.yaml to verify it is excluded.

​
Hot-Reload Behavior

ADE watches all three config files using filesystem events (debounced 500ms). Changes take effect without restarting ADE.
FileWhat happens on change
ade.yamlADE detects the change, diffs against the approved SHA, and suspends commands until you approve the new version
local.yamlApplied immediately; UI reflects changes (theme, font, notification preferences) within one render cycle
local.secret.yamlAPI key changes take effect on the next agent call; MCP server credential changes trigger reconnection of the affected server
If you update an API key in local.secret.yaml while an agent is mid-session, the new key is used on the next API call. The in-flight request completes with the old key. No session restart is needed.

​
Best Practices

Commit ade.yaml. It contains the team’s shared AI model defaults, lane profiles, automation definitions, test suite config, and process definitions. Committing it ensures all team members use the same baseline configuration.Do not commit local.yaml or local.secret.yaml. ADE’s .gitignore entries cover these, but double-check when initializing new repositories.
To rotate an API key: update the value in local.secret.yaml and save. ADE hot-reloads the file and uses the new key on the next API call. The old key is never cached — once the file is saved, the old value is gone from memory.Never commit the old or new key values to git during rotation. If a key was accidentally committed, revoke it at the provider immediately regardless of whether you have since removed it from the file.
When you git pull and ade.yaml has changed, ADE shows the approval dialog before resuming any command execution. Read the diff carefully — pay particular attention to changes in automations[*].executor, processes[*].command, and testSuites[*].ci. These fields define commands that will run on your machine.
For CI environments or machines where you cannot write files, ADE supports supplying API keys via environment variables. Set ADE_ANTHROPIC_API_KEY, ADE_OPENAI_API_KEY, etc. in the process environment before launching ADE. Environment variable values take precedence over local.secret.yaml.
Each project has its own .ade/ directory with its own local.yaml and local.secret.yaml. You can use different API keys or different preferred models for different projects. Settings in one project’s .ade/ never affect another project.
Screenshot: The Settings panel showing the three config files, the Approved SHA display, and the Re-approve button.

​
What’s Next

AI Providers

Configure Anthropic, OpenAI, OpenRouter, and local models. Set budgets and context preferences.

MCP Servers

Connect external MCP servers like GitHub, Slack, and Ghost OS to give agents additional tools.

Settings

Full reference for every setting in ADE’s Settings page.

Permissions

Understand ADE’s trust boundary model and configure agent permissions.