fabro/docs/administration/advanced-setup.mdx

---
title: "Advanced Setup"
description: "Server configuration, environment variables, production deployment, and troubleshooting"
---

## Server configuration

The server config file at `~/.arc/server.toml` controls how `arc serve` behaves — API binding, authentication, run defaults, and more. The [Quick Start](/getting-started/quick-start) doesn't require one, but production deployments should configure it explicitly.

### Full reference

```toml
# Maximum concurrent workflow runs (default: 5)
max_concurrent_runs = 8

# Override the default data directory (default: ~/.arc)
data_dir = "/var/lib/arc"

[api]
base_url = "https://arc.example.com"

[api.tls]
cert = "/etc/arc/tls/cert.pem"
key = "/etc/arc/tls/key.pem"
ca = "/etc/arc/tls/ca.pem"

# Authentication strategies (array of Jwt or Mtls)
[[api.authentication_strategies]]
type = "Jwt"

[web]
url = "https://arc-web.example.com"

[web.auth]
provider = "Github"
allowed_usernames = ["alice", "bob"]

[git]
provider = "Github"
app_id = "123456"
client_id = "Iv1.abc123"

# Run defaults — applied to every run unless overridden by the run config
[llm]
model = "claude-sonnet-4-5"
provider = "anthropic"

[llm.fallbacks]
anthropic = ["gemini", "openai"]

[setup]
commands = ["npm install"]
timeout_ms = 120000

[sandbox]
provider = "daytona"

[sandbox.daytona]
auto_stop_interval = 60

[sandbox.daytona.labels]
team = "platform"

[vars]
default_branch = "main"
```

### CLI overrides

Several `server.toml` settings can be overridden via `arc serve` flags:

| Flag | Default | Description |
|---|---|---|
| `--port` | `3000` | Port to listen on |
| `--host` | `127.0.0.1` | Host address to bind to |
| `--model` | — | Override default LLM model |
| `--provider` | — | Override default LLM provider |
| `--sandbox` | — | Override default sandbox provider |
| `--max-concurrent-runs` | `5` | Maximum concurrent run executions |
| `--config` | `~/.arc/server.toml` | Path to server config file |
| `--dry-run` | — | Execute with simulated LLM backend |

CLI flags take precedence over `server.toml` values. See [Run Configuration — Precedence](/execution/run-configuration#precedence) for the full resolution order.

### Run defaults

The `[llm]`, `[setup]`, `[sandbox]`, and `[vars]` sections in `server.toml` act as defaults for every run. A run config TOML can override any of these. For `[vars]` and Daytona labels, values are **merged** — the run config wins on key collisions. All other fields use "first non-empty wins" precedence.

## Environment variables

Arc reads environment variables from a `.env` file in the working directory (if present) and from the shell environment. Provider API keys are required for the models you want to use; everything else is optional.

### LLM provider keys

| Variable | Provider |
|---|---|
| `ANTHROPIC_API_KEY` | Anthropic (Claude) |
| `OPENAI_API_KEY` | OpenAI (GPT) |
| `GEMINI_API_KEY` or `GOOGLE_API_KEY` | Google (Gemini) |
| `KIMI_API_KEY` | Kimi |
| `ZAI_API_KEY` | Zai (GLM) |
| `MINIMAX_API_KEY` | Minimax |
| `INCEPTION_API_KEY` | Inception (Mercury) |

### Sandbox and tools

| Variable | Description |
|---|---|
| `DAYTONA_API_KEY` | Daytona cloud sandbox API key |
| `BRAVE_SEARCH_API_KEY` | Brave Search API key (for the `web_search` tool) |

### Server authentication

| Variable | Description |
|---|---|
| `ARC_JWT_PRIVATE_KEY` | Ed25519 private key (base64-encoded PEM) for JWT signing |
| `ARC_JWT_PUBLIC_KEY` | Ed25519 public key (base64-encoded PEM) for JWT verification |
| `SESSION_SECRET` | Session encryption secret (64-character hex string) |

### GitHub App (optional)

| Variable | Description |
|---|---|
| `GITHUB_APP_CLIENT_SECRET` | GitHub App client secret |
| `GITHUB_APP_WEBHOOK_SECRET` | GitHub App webhook secret |
| `GITHUB_APP_PRIVATE_KEY` | GitHub App private key (base64-encoded) |

### Slack integration (optional)

| Variable | Description |
|---|---|
| `ARC_SLACK_APP_TOKEN` | Slack App-level token |
| `ARC_SLACK_BOT_TOKEN` | Slack Bot token |

### Logging

| Variable | Default | Description |
|---|---|---|
| `ARC_LOG` | `info` | Log level: `error`, `warn`, `info`, `debug` |

## Production deployment

### Binding and TLS

By default, `arc serve` binds to `127.0.0.1:3000` (localhost only). For production, either:

1. **Reverse proxy** — Keep the default binding and place nginx, Caddy, or an ALB in front:

```nginx
upstream arc {
    server 127.0.0.1:3000;
}

server {
    listen 443 ssl;
    server_name arc.example.com;

    ssl_certificate /etc/tls/cert.pem;
    ssl_certificate_key /etc/tls/key.pem;

    location / {
        proxy_pass http://arc;
        proxy_http_version 1.1;

        # Required for SSE event streams
        proxy_set_header Connection '';
        proxy_buffering off;
        proxy_cache off;
    }
}
```

2. **Direct TLS** — Configure TLS in `server.toml`:

```toml
[api.tls]
cert = "/etc/arc/tls/cert.pem"
key = "/etc/arc/tls/key.pem"
ca = "/etc/arc/tls/ca.pem"
```

<Note>
If you use a reverse proxy, set `api.base_url` in `server.toml` to the external URL so that API responses include correct links.
</Note>

### Running as a service

Example systemd unit:

```ini
[Unit]
Description=Arc API Server
After=network.target

[Service]
Type=simple
User=arc
WorkingDirectory=/opt/arc
EnvironmentFile=/opt/arc/.env
ExecStart=/opt/arc/bin/arc serve --host 0.0.0.0 --port 3000 --config /etc/arc/server.toml
Restart=on-failure
RestartSec=5

[Install]
WantedBy=multi-user.target
```

### Health checks

The API server exposes a health endpoint at `GET /health` that returns `{"status": "ok"}`. No authentication is required. Use this for load balancer health checks, container orchestration probes, or uptime monitoring.

```bash
curl http://localhost:3000/health
```

### Concurrency tuning

| Setting | Default | Where |
|---|---|---|
| Max concurrent runs | 5 | `--max-concurrent-runs` or `max_concurrent_runs` in `server.toml` |
| Max parallel branches per node | 4 | `max_parallel` attribute on parallel nodes in DOT |
| Subagent depth | 1 | Hardcoded (subagents can spawn one level deep) |
| Max tool rounds per turn | 200 | Hardcoded |
| Default retries | 3 | `default_max_retry` graph attribute (or `max_retries` per node) |

### Stall watchdog

Arc cancels a run if no handler events are emitted for a configurable duration. This protects against hung agents or unresponsive LLM providers.

| Setting | Default | Where |
|---|---|---|
| `stall_timeout` | `600s` (10 min) | Graph attribute: `graph [stall_timeout="300s"]` |

Set to `0s` to disable: `graph [stall_timeout="0s"]`.

### Log files

Arc writes two kinds of logs:

- **Application logs** — `~/.arc/logs/YYYY-MM-DD.log` (structured, controlled by `ARC_LOG`)
- **Run logs** — `~/.arc/logs/{run_id}/progress.jsonl` (one event per line)

The logs directory is not currently configurable and has no automatic retention. For production, set up external log rotation (e.g. `logrotate`) or periodic cleanup of old run directories.

See [Observability](/execution/observability) for the full event schema and analysis tools.

## Troubleshooting

### `arc doctor`

The `arc doctor` command validates your installation:

```bash
arc doctor           # Check local configuration
arc doctor --live    # Also probe live services (LLM APIs, sandbox, Brave Search)
arc doctor --verbose # Show detailed output for each check
```

It checks:
- System dependencies (`openssl`, `node`, `gh`, `dot`)
- LLM provider API keys
- Sandbox availability (Docker daemon, Daytona API key)
- JWT key configuration
- Brave Search API key

### Common issues

**"No API key configured"** — Set at least one provider key in `.env` or your shell environment. Run `arc doctor --live` to verify connectivity.

**Stall watchdog timeouts** — If runs are cancelled unexpectedly, the agent may be stuck or the LLM provider may be slow. Check `ARC_LOG=debug` output for `Agent.LlmRetry` events. Increase `stall_timeout` in the graph if needed, or add [fallback providers](/core-concepts/models) to handle outages.

**Sandbox creation failures** — For Docker: ensure the Docker daemon is running and the configured image exists. For Daytona: verify `DAYTONA_API_KEY` is set and the `gh` CLI is authenticated.

**Port already in use** — Change the port with `arc serve --port 3001` or stop the conflicting process.

**SSE streams disconnecting** — If using a reverse proxy, ensure buffering is disabled and the connection timeout is long enough for workflow runs. See the nginx example above.

**Run config validation errors** — Use `--preflight` to validate without executing:

```bash
arc run start run.toml --preflight
```