Configuration

Configuration is loaded from user scope and project scope, with project values overriding user values.

Paths (see Paths for platform-specific details):

• User: config dir + config.toml
• Project: <cwd>/.acolyte/config.toml

Common commands

acolyte config list
acolyte config set model gpt-5-mini
acolyte config set locale en
acolyte config set --project logFormat json
acolyte config unset openaiBaseUrl

Vercel AI Gateway (recommended)

The fastest way to get started. The Vercel AI Gateway provides unified access to 20+ providers with a single API key.

acolyte init vercel
acolyte config set model anthropic/claude-sonnet-4

When a direct provider key is also set (e.g. ANTHROPIC_API_KEY), Acolyte prefers the direct connection. When it’s missing, requests fall back to the gateway automatically — no prefix or config change needed.

# Explicitly target a provider only available through the gateway
acolyte config set model vercel/xai/grok-4.1

# Override the gateway base URL
acolyte config set vercelBaseUrl https://custom-gateway.example.com/v1

Provider base URLs

Each provider has a configurable base URL with a sensible default:

• openaiBaseUrl: OpenAI API base (default: https://api.openai.com/v1). Set to a local endpoint for OpenAI-compatible providers (Ollama, vLLM, etc.).
• anthropicBaseUrl: Anthropic API base (default: https://api.anthropic.com/v1). Must end with /v1.
• googleBaseUrl: Google AI API base (default: https://generativelanguage.googleapis.com).

Local models

Configure an OpenAI-compatible local endpoint directly in project config, then set the model explicitly:

acolyte config set --project openaiBaseUrl http://localhost:11434/v1
ollama pull <model>
acolyte config set --project model openai-compatible/<model>

Localization

• locale: active UI language (defaults to en).
• English messages are defined in src/i18n/en.ts. Additional locales are loaded from src/i18n/locales/*.json at startup.

Logging

• logFormat: log output format (logfmt | json, default: logfmt).

logfmt emits one key=value line per entry:

2026-03-20T12:00:00.000Z level=info msg="request started" model=gpt-5-mini

json emits one JSON object per line with typed fields:

{"ts":"2026-03-20T12:00:00.000Z","level":"info","msg":"request started","model":"gpt-5-mini"}

acolyte config set logFormat json

MCP servers

Connect Acolyte to external services (Figma, Jira, Notion, Chrome DevTools, etc.) via MCP servers. Configure servers under the mcpServers key in .mcp.json at the project root.

MCP is disabled by default. Enable it explicitly with:

[features]
mcp = true

Only enable MCP in trusted repositories. .mcp.json is project-controlled and stdio servers execute local commands.

Two transports are supported:

• stdio — server runs as a local subprocess (command, args, optional env). Only a minimal set of environment variables is forwarded to the subprocess (PATH, HOME, SHELL, TERM, USER, LANG, LC_ALL, TMPDIR, XDG_RUNTIME_DIR) plus any explicitly configured in env.
• http — server is reachable over HTTPS (url, optional headers). Non-HTTPS URLs are allowed for localhost (127.0.0.1, ::1) but blocked for remote hosts.

Each server’s tools appear in the agent alongside native tools. If a server is unreachable at task start it is skipped with a warning and the lifecycle continues.

Feature flags

Feature flags are opt-in toggles for experimental behavior, configured under [features] in config.toml.

Enable via TOML:

[features]
syncAgents = true

Enable via CLI:

acolyte config set features.syncAgents true

Available flags

All settable keys