funky

Manage Funky Cloud agents as code. Describe each agent in a YAML file, then funky apply creates or updates it — like Terraform, for agent configs.

Install

npm install -g @funky-dev/cli    # or, without installing: npx @funky-dev/cli login

Or download a binary for macOS, Linux or Windows from the latest release, or, with Go installed:

go install github.com/funkyhq/cli/cmd/funky@latest

funky --version shows the installed version.

Quick start

funky login      # opens your browser: sign in, pick a project, approve this machine
funky init       # creates agents/
$EDITOR agents/researcher.yaml
funky apply      # shows the plan, asks, then creates or updates

Agent files

One file per agent config in agents/. The fields are the API's agent config fields:

# agents/researcher.yaml
inference:
  provider: anthropic
  model: claude-sonnet-5
  maxTokens: 2048
systemPrompt: You are an autonomous agent in a fresh Linux sandbox.
network:
  type: allowlist
  domains: [github.com]
packages:
  pip: [pandas==2.2.0]
setupCommands: []
credentials: []
resources: { cpuCount: 4, memoryMB: 4096 }
metadata: { team: research }

inference, systemPrompt and network are required; the rest default to empty. resources sizes the agent's sandbox in vCPUs and MiB of memory; a field left out is the default, 2 vCPUs and 1024 MiB.

An agent is named by its file name, or by a name: field. The name is stored in the config's metadata as funky.dev/name, which is how the next apply finds the config to update.

funky apply:

• creates configs for new files, updates configs whose file changed (sending only the changed fields, so each apply adds at most one version), and leaves the rest alone;
• pins each update to the version it planned against, so a config edited elsewhere in the meantime fails with a conflict instead of being overwritten;
• never archives anything: removing a file leaves its config in place.

Flags: --dry-run shows the plan only; --auto-approve skips the confirmation.

Authentication

funky login signs the CLI in as you, in one project — run it again to switch projects. The login lasts 90 days, lives in your OS keychain, and can be revoked with funky logout or from Settings in the web app. funky whoami shows who and where the CLI is acting.

On a machine without a browser, funky login --no-browser prints a link and asks for the code the page shows.

In CI, set FUNKY_API_KEY to a key from https://platform.funky.dev/api-keys instead; it takes precedence over a login.

Development

go test ./...
go run ./cmd/funky --help

FUNKY_API_URL, FUNKY_CONSOLE_URL and FUNKY_PLATFORM_URL point the CLI at non-production services; FUNKY_CONFIG_DIR moves where the login is saved.

Releasing

Push a version tag; the release workflow builds every platform with GoReleaser, publishes a GitHub release, then publishes @funky-dev/cli to npm (npm/publish.mjs; each platform's binary is its own version, like 0.1.1-darwin-arm64, which the main version installs as an optional dependency):

git tag v0.1.0 && git push origin v0.1.0

A tag with a suffix (v0.2.0-rc.1) becomes a pre-release, on npm's next tag. To try the build locally: goreleaser release --snapshot --clean && node npm/publish.mjs --dry-run (output in dist/).