fledge

One CLI for the dev loop. Any language. JSON by default. Read the docs and go.

fledge templates init my-tool --template rust-cli
cd my-tool
fledge lanes init
fledge lanes run ci

Working with AI agents? See AGENTS.md. Every command emits {schema_version: 1, ...}, FLEDGE_NON_INTERACTIVE=1 silences prompts, fledge ask and fledge review are spec-aware, and fledge introspect --json dumps the full command tree. Talks to the Anthropic API, any OpenAI-compatible endpoint (OpenAI, OpenRouter, Groq, ...), or any Ollama endpoint (local, cloud, or self-hosted) over plain HTTP. No CLI required.

Install

cargo install fledge                       # crates.io
brew install CorvidLabs/tap/fledge         # homebrew

fledge plugins install --defaults          # curated plugin set

More install options

curl -fsSL https://raw.githubusercontent.com/CorvidLabs/fledge/main/install.sh | sh
nix run github:CorvidLabs/fledge
git clone https://github.com/CorvidLabs/fledge.git && cd fledge && cargo install --path .

GitHub Actions

- uses: CorvidLabs/fledge@v1
  with:
    version: v1.7.2

Pin version to a release tag and the action downloads that binary directly — no GitHub API call, so nothing to rate-limit on shared runner IPs. Leave it at the default latest and the action resolves the tag via an authenticated API call (using github.token unless you pass your own token). version accepts a release tag or latest and nothing else, since the value reaches a download URL.

Every download is verified against the release's .sha256 sidecar, and a checksum that is missing, unfetchable, or mismatched fails the step — the install never degrades to an unverified binary. Releases before v0.9.1 predate the sidecars and so cannot be installed this way.

Linux and macOS runners (x86_64/aarch64) are supported. Windows is not supported by the action yet, though fledge does publish fledge-windows-x86_64.exe in every release for direct download.

outputs:
  version: # the release tag that was installed
  path:    # full path to the installed binary

Quick start

Already have a project? cd into it, fledge auto-detects the stack:

fledge run test       # runs your language's test command
fledge run build      # same for build
fledge run test -- --release    # pass extra args through to the command after --
fledge review         # AI code review against the default branch
fledge review --with-model ollama:gpt-oss:120b-cloud,ollama:qwen3-coder:480b-cloud
              # multi-model panel, parallel critiques on the same diff

Starting fresh? Scaffold from a template:

fledge templates init my-app --template rust-cli     # built-in
fledge templates init my-app --template user/repo    # any GitHub repo
fledge templates init my-app                         # interactive picker

Switch AI providers without editing config:

fledge ai use                                  # interactive picker (live Ollama model list)
fledge ai use ollama qwen3-coder:480b-cloud    # scriptable
fledge ai status                               # show active provider/model and where each value came from

The six pillars

Pillar	Commands	What it does
Scaffold	templates (init, create, list, search, validate, publish)	Start a project from a template, local or remote
Run	run, lanes, watch	Task runner, composable pipelines, file-watch reruns
Spec	spec	spec-sync. Modules declare their contract, AI uses it as context
AI	ai, ask, review	Provider/model selection, spec-aware Q&A, single and multi-model code review
Ship	work, release, changelog	Branch and PR flow with AI-drafted bodies, version bump, tag, push
Extend	plugins, config, introspect, completions, doctor	Plugin protocol, global config, command-tree introspection, env health

That is the whole core. Anything else is a plugin.

Plugins

Plugins extend fledge with community-built commands. Native plugins run as regular executables. WASM plugins run in a sandboxed Wasmtime runtime with no host access by default.

fledge plugins install --defaults          # curated native plugin set
fledge plugins create my-lint --wasm       # scaffold a sandboxed WASM plugin

Default plugins

Three native plugins ship as the default set:

Plugin	Adds
fledge-plugin-github	checks, issues, prs. GitHub PR/issue/CI flow
fledge-plugin-deps	deps. Polyglot lockfile audits
fledge-plugin-metrics	metrics. LOC, churn, test/source ratio (via tokei + git)

WASM plugins

WASM plugins are ideal for pure-computation tasks (linting, formatting, analysis) where you want strong isolation without trusting arbitrary binaries:

• Sandboxed by default. No filesystem, no network
• Opt-in capabilities prompted at install time
• Fuel-bounded execution (no infinite loops)
• 256 MB memory cap
• Cross-platform single .wasm binary

See the WASM plugin guide for authoring details.

Built-in templates

Language starters: rust-cli, ts-bun, python-cli, go-cli, ts-node, static-site, kotlin-kmp, kotlin-ktor-api

Setup-only: corvid-stack (CorvidLabs trust toolchain — fledge + spec-sync + augur + attest, no language scaffolding), fledge-plugin (plugin scaffold)

Browse community templates: fledge templates search <keyword>

Examples

• Community templates. A growing collection covering Angular, Bun APIs, Deno CLIs, MCP servers, Rust workspaces, Swift packages, and more
• Example lanes. Language-specific CI/CD pipelines
• Example plugin. Deploy/rollback plugin reference

Learn more

• Full documentation. Commands, configuration, guides
• Template authoring. How to create and publish your own templates
• Lanes guide. Task pipelines and workflow automation
• Plugins guide. Extend fledge with community tools
• WASM plugins. Build sandboxed plugins with Wasmtime

Contributing

See CONTRIBUTING.md for development setup, guidelines, and how to submit changes.

Security

See SECURITY.md for the security policy and how to report vulnerabilities.

License

MIT