Documentation

Configuration

Quick Setup

Use a preset to get started fast:

# CorvidLabs preset - sets author, org, license, template repo
fledge config init --preset corvidlabs

# Default config
fledge config init

Config File

Lives under the platform config directory:

Platform	Path
macOS	`~/Library/Application Support/fledge/config.toml`
Linux	`~/.config/fledge/config.toml`
Windows	`%APPDATA%\fledge\config.toml`

Sections

[defaults]

Default values for new projects:

[defaults]
author = "Your Name"
github_org = "YourOrg"
license = "MIT"

Key	What it does	Fallback
`author`	Default author name	`git config user.name`
`github_org`	Default GitHub org	Prompted
`license`	Default license	`MIT`

[templates]

Where to find templates:

[templates]
paths = ["~/my-templates", "~/work/templates"]
repos = ["CorvidLabs/fledge-templates", "myorg/templates"]

Key	What it does
`paths`	Local directories with templates
`repos`	GitHub repos to pull templates from (`owner/repo`)

[ai]

AI provider and model settings. Written by fledge ai use or fledge config set/edit:

All providers are reached over plain HTTP (no CLI). provider is one of ollama, anthropic, openai, openrouter, gemini, deepseek, groq, mistral, xai, together (claude is a deprecated alias of anthropic). When unset, fledge auto-detects the first provider with a key, falling back to keyless local Ollama.

[ai]
provider = "anthropic"          # ollama | anthropic | openai | openrouter | gemini | deepseek | groq | mistral | xai | together

[ai.anthropic]
model = "claude-sonnet-4-6"     # default if unset
api_key = "sk-ant-..."          # or export ANTHROPIC_API_KEY
# base_url = "..."              # optional override

[ai.openai]                     # any OpenAI-compatible gateway
base_url = "https://openrouter.ai/api/v1"
api_key = "sk-..."              # or export OPENAI_API_KEY
model = "anthropic/claude-sonnet-4-6"  # required (no default)

[ai.ollama]
host = "http://localhost:11434" # Ollama API endpoint (always normalized to include scheme)
model = "llama3.3"              # use `fledge ai models --provider ollama` to list available models
api_key = "sk-..."             # for Ollama Cloud / authenticated endpoints
timeout_seconds = 600          # request timeout (default: 600)

The OpenAI-compatible gateways (openrouter, groq, deepseek, mistral, xai, together, gemini) are key-driven: set <PROVIDER>_API_KEY (e.g. GROQ_API_KEY); their endpoint and default model come from the corvid-ai registry.

Tip: Run fledge ai models --provider ollama (live) or --provider anthropic (curated) to see models. Use fledge ai use for an interactive picker.

Key	What it does	Default
`ai.provider`	Active LLM backend	auto-detect → keyless local Ollama
`ai.anthropic.model`	Anthropic model id	`claude-sonnet-4-6`
`ai.anthropic.api_key`	Anthropic key (or `ANTHROPIC_API_KEY`)	(none)
`ai.anthropic.base_url`	Anthropic endpoint override	`https://api.anthropic.com`
`ai.openai.base_url`	OpenAI-compatible gateway URL	OpenAI
`ai.openai.api_key`	OpenAI-compatible key (or `OPENAI_API_KEY`)	(none)
`ai.openai.model`	Model id (required: no default)	(none)
`ai.ollama.host`	Ollama API endpoint URL	`http://localhost:11434`
`ai.ollama.model`	Ollama model name	`llama3.3`
`ai.ollama.api_key`	Bearer token for authenticated endpoints	(none)
`ai.ollama.timeout_seconds`	Request timeout in seconds	`600`

[trust]

Extend the plugin/lane trust system without recompiling. Entries here classify as team tier, granting access to exec and network capabilities:

[trust]
orgs = ["my-company", "my-other-org"]
users = ["my-friend", "trusted-bot"]

Key	What it does
`orgs`	GitHub orgs to trust at team tier
`users`	GitHub users to trust at team tier

Manage via CLI:

fledge config add trust.orgs my-company
fledge config add trust.users 0xLeif
fledge config remove trust.orgs my-company

Compared case-insensitively. The hardcoded official org (CorvidLabs) and built-in team members are always trusted regardless of config.

[github]

[github]
token = "ghp_..."

Token priority:

FLEDGE_GITHUB_TOKEN env var
GITHUB_TOKEN env var
Config file
gh auth token (GitHub CLI fallback)

Required token scopes:

Feature	Scopes needed
Issues, PRs, CI checks	`repo` (or `public_repo` for public repos only)
Create PRs, push branches	`repo`
Search templates/plugins	`public_repo`
Publish templates	`repo`, `delete_repo` (if republishing)

A classic token with repo covers everything. For fine-grained tokens, grant Read/Write on Contents, Pull Requests, and Issues for each repo you work with.

Full Example

[defaults]
author = "Leif"
github_org = "CorvidLabs"
license = "MIT"

[templates]
paths = ["~/.fledge/templates", "~/projects/templates"]
repos = ["CorvidLabs/fledge-templates", "my-org/my-templates"]

[trust]
orgs = ["my-company"]
users = ["0xLeif"]

[github]
token = "ghp_..."

[ai]
provider = "anthropic"

[ai.anthropic]
model = "claude-sonnet-4-6"

[ai.ollama]
host = "http://localhost:11434"
model = "llama3.3"
timeout_seconds = 600

Environment Variables

Variable	What it does
`FLEDGE_NON_INTERACTIVE`	Truthy (`1`, `true`, `yes`, `y`, `on`) silences prompts. Same effect as passing `--non-interactive` (alias `--ni`) per invocation. Confirmation prompts behave as `--yes`; prompts with no default bail with a clear error instead of hanging. Hook exception: for ad-hoc remote templates passed to `templates init` (`--template owner/repo` not in config), `post_create` hooks are skipped unless trust is granted explicitly via `--trust-hooks` or `FLEDGE_TRUST_HOOKS`. Built-in templates, anything under `templates.paths`, and anything reachable through your configured `templates.repos` still follow the `--yes` consent path.
`FLEDGE_TRUST_HOOKS`	Truthy authorizes `post_create` hook execution for ad-hoc remote templates passed to `fledge templates init`. Same as passing `--trust-hooks`. Has no effect on templates reached through `templates.paths` or `templates.repos` (those already follow the `--yes` consent path). Hooks run arbitrary shell commands; only set this for sources you trust.
`FLEDGE_GITHUB_TOKEN`	GitHub token (highest priority)
`GITHUB_TOKEN`	GitHub token (fallback after FLEDGE_GITHUB_TOKEN)
`FLEDGE_AI_PROVIDER`	AI provider override (`ollama`, `anthropic`, `openai`, `openrouter`, `gemini`, `deepseek`, `groq`, `mistral`, `xai`, `together`)
`FLEDGE_AI_MODEL`	AI model override
`FLEDGE_AI_TIMEOUT`	Request timeout in seconds
`ANTHROPIC_API_KEY`	Anthropic API key
`OPENAI_API_KEY`	OpenAI-compatible API key
`<PROVIDER>_API_KEY`	Gateway key, e.g. `GROQ_API_KEY`, `GEMINI_API_KEY`, `OPENROUTER_API_KEY`
`OLLAMA_HOST`	Ollama API endpoint URL
`OLLAMA_API_KEY`	Ollama Bearer token

If neither env var nor config is set, fledge falls back to gh auth token (GitHub CLI) automatically for GitHub operations.

Project Configuration (fledge.toml)

Per-project settings live in fledge.toml in your project root. This file defines tasks, lanes, and release behavior. It's created by fledge run --init, fledge lanes init, or fledge templates init.

For the full schema (every section, every key, every default), see the fledge.toml Reference.

Topical guides:

Run: Tasks and Lanes. Defining tasks/lanes, step types, parallel groups, importing community lanes
Extend: Plugins. plugin.toml and the plugin ecosystem

Priority Order

When creating a project, values come from (highest to lowest):

Command-line arguments
Config file
Git config (author only)
Built-in defaults