fusion/packages/cli/STANDALONE.md

Standalone CLI

Fusion works as a standalone CLI without pi. This is useful for CI environments, scripting, or if you prefer working from the terminal.

Installation

npm install -g @runfusion/fusion

Authentication

Fusion uses pi for AI agent sessions and reuses your existing pi authentication. You can also authenticate directly through the dashboard UI.

If you don't have pi set up yet: npm i -g @mariozechner/pi-coding-agent && pi then /login.

Usage

Start the dashboard

Launch the web UI and AI engine:

fn dashboard
fn dashboard --port 8080
fn dashboard --interactive     # Interactive port selection (prompts for port)
fn dashboard --paused        # Start with automation paused (review before work begins)
fn dashboard --dev           # Start web UI only (no AI engine)

Multi-Instance Deployments

When deploying the dashboard behind a load balancer with multiple instances, configure Redis pub/sub for real-time badge updates across instances:

# Set Redis URL for cross-instance badge synchronization
export FUSION_BADGE_PUBSUB_REDIS_URL="redis://redis.example.com:6379"

# Optional: customize the pub/sub channel (default: fusion:badge-updates)
export FUSION_BADGE_PUBSUB_CHANNEL="my-app-badge-updates"

fn dashboard

With this configuration, PR/issue badge updates received via webhook on one instance are delivered to subscribed WebSocket clients on all instances.

GitHub App Webhook Setup

For real-time PR/issue badge updates, configure a GitHub App to push updates to the dashboard:

Required Environment Variables:

export FUSION_GITHUB_APP_ID="123456"
export FUSION_GITHUB_APP_PRIVATE_KEY_PATH="/path/to/private-key.pem"
# Or: export FUSION_GITHUB_APP_PRIVATE_KEY="$(cat /path/to/private-key.pem)"
export FUSION_GITHUB_WEBHOOK_SECRET="your-webhook-secret"

GitHub App Configuration:

1. Create a GitHub App at Settings → Developer settings → GitHub Apps
2. Set the Webhook URL to https://your-domain/api/github/webhooks
3. Generate and download a Private Key
4. Configure these Permissions:
   • Metadata: Read
   • Pull requests: Read
   • Issues: Read
5. Subscribe to these Webhook Events:
   • Pull request
   • Issues
   • Issue comment

Minimum Permissions Summary:

Permission	Level	Purpose
Metadata	Read	Access repository metadata
Pull requests	Read	Fetch PR status, title, comments
Issues	Read	Fetch issue status, title, state

Fallback Behavior: When webhooks are not configured or delivery fails, the dashboard falls back to the 5-minute background refresh on the PR/issue status endpoints. The 5-minute staleness window ensures reasonably fresh data even without webhooks.

Create a task

fn task create "Fix the login redirect bug"
fn task create "Update hero section" --attach screenshot.png --attach design.pdf

Manage tasks

fn task list                        # List all tasks
fn task show KB-001                 # Show task details, steps, and log
fn task move KB-001 todo            # Move a task to a column
fn task merge KB-001                # Merge an in-review task and close it
fn task pr-create KB-001            # Create a GitHub PR for an in-review task
fn task pr-create KB-001 --title "Custom PR title" --base develop --body "PR description"
fn task log KB-001 "Added context"  # Add a log entry
fn task pause KB-001                # Pause a task (stops automation)
fn task unpause KB-001              # Resume a paused task
fn task attach KB-001 ./error.log   # Attach a file to a task
fn task import owner/repo           # Import GitHub issues as tasks
fn task import owner/repo --limit 10 --labels "bug,enhancement"

Agent control

fn agent stop <agent-id>            # Stop (pause) a running agent
fn agent start <agent-id>           # Start (resume) a stopped agent
fn agent mailbox <agent-id>         # View an agent's mailbox

Messaging

fn message inbox                    # List inbox messages
fn message outbox                   # List sent messages
fn message send <agent-id> <msg>    # Send a message to an agent
fn message read <id>                # Read a specific message
fn message delete <id>              # Delete a message

Typical workflow

# 1. Create a task — it lands in triage
fn task create "Add dark mode support"

# 2. Start the dashboard — AI specs the task and begins working
fn dashboard

# 3. Check progress
fn task list
fn task show KB-042

# 4. When it reaches "in-review", review the changes and merge
fn task merge KB-042

Standalone binary

Prebuilt standalone binaries are available that require no Node.js runtime. You can also build one yourself with Bun:

bun run build.ts

Runtime Assets

When using standalone binaries, the dashboard's integrated terminal requires native platform assets that must be co-located with the binary:

dist/
├── kb                    # Binary (or kb-darwin-arm64, kb-linux-x64, etc.)
├── client/               # Dashboard web assets (required)
└── runtime/              # Native terminal assets (required for terminal)
    └── darwin-arm64/     # Platform-specific subdirectory
        ├── pty.node      # Native PTY module
        └── spawn-helper  # Unix spawn helper (macOS/Linux only)

Platform-specific subdirectories:

• darwin-arm64/ - macOS Apple Silicon
• darwin-x64/ - macOS Intel
• linux-arm64/ - Linux ARM64
• linux-x64/ - Linux x64
• win32-x64/ - Windows x64

Important: When distributing or moving the binary, ensure the client/ and runtime/ directories are copied alongside it. Terminal functionality will gracefully degrade (return HTTP 503) if runtime assets are missing — the dashboard will continue to work but terminal sessions won't be available.

How it works: When the dashboard starts from a Bun-compiled binary, it attempts to set up native module resolution so node-pty can find its platform-specific .node files. This involves:

1. Copying native assets to a temp directory (/tmp/kb-bunfs-<pid>/kb/prebuilds/<platform>/)
2. Attempting to create a symlink at /$bunfs/root pointing to the temp directory (Unix platforms)
3. If the symlink can't be created (e.g., macOS permissions), pre-loading the native module via process.dlopen()

If all resolution methods fail, terminal creation gracefully returns null, which the HTTP layer converts to a 503 Service Unavailable response.

Cross-compilation: Native assets are staged per-platform during build. When cross-compiling, only the target platform's assets are included. PTY functionality requires running on a platform with matching native assets.

Known Bun node:sqlite Limitation

Bun-compiled standalone binaries may encounter a No such built-in module: node:sqlite error at startup. This happens because Bun's compiler does not include the full node:sqlite built-in module in all compilation targets.

Impact: When this error occurs, the binary exits immediately. This affects any command that initializes the SQLite-backed task store, including dashboard, task list, and task create. Commands that don't need the store (like --help) continue to work.

Detection: The startup validation test suite treats this specific error as an expected limitation — it is not misinterpreted as a generic dashboard startup failure. The test probe distinguishes between:

Outcome	Behavior
Startup banner detected	Full test proceeds (PTY endpoint verification)
node:sqlite error in output	Test skips cleanly (known Bun limitation)
Other early exit	Test fails with diagnostic output

Only the exact node:sqlite built-in module error is handled specially. Any other exit or crash during startup is treated as a real regression and fails the test with full process output for debugging.