Skip to content
dbowm91Public

About

AI harness for hatching new code

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Latest commit

 

History

4,789 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

codegg

CodeGG is a Rust-native AI coding agent for terminal workflows: an interactive TUI, persistent sessions, multiple LLM providers, and code/shell tooling behind a permission gate. It is at 0.1.0 and under active development.

Quick start

You need one LLM provider credential and network access to it.

1. Install

git clone https://github.com/dbowm91/codegg.git
cd codegg
cargo install --locked --path .

No GitHub release has been published yet, so there is no prebuilt installer. cargo install also installs codegg-sandbox-helper. The managed codegg-eggsearch sidecar (web/repository search) is not built from this repo; codegg doctor reports whether it is present and how to point CodeGG at one. Other paths are in docs/install.md.

2. Add a provider key

export ANTHROPIC_API_KEY='...'
codegg providers   # what your config can actually reach

codegg providers and codegg models -p anthropic read your live configuration — entries appear only when their credential resolves. Full credential options are in docs/providers.md.

3. Pick a model

With no model configured, CodeGG falls back to openai/gpt-4o, so set one:

{
  "model": "anthropic/claude-sonnet-4-20250514"
}

or pass -m provider/model-id per invocation. Validate with codegg validate (codegg validate --config codegg.example.jsonc checks the annotated example). Config file locations and the full schema are in docs/configuration.md.

4. Check the setup

codegg doctor   # what is wired up and what is missing (exit 0 even with problems)

5. Run it

codegg                                # new session in the current directory
codegg -c                             # resume the last session
codegg -m anthropic/<model-id>        # pick a model
codegg --cwd /path/to/project         # choose the workspace (must exist)
codegg --run "Explain this project"   # one prompt, then exit

/connect in the TUI walks you through provider onboarding on a clean install. Every flag is in docs/cli.md.

Example usage

One-shot and CI (codegg exec reads {"prompt","model","agent"} from --json, --file, or stdin; a finished run prints one JSON object and exits non-zero on failure — startup failures such as an unknown provider report a plain error on stderr instead):

printf '%s\n' '{"prompt":"Summarize the changes in this repository","model":"anthropic/claude-sonnet-4-20250514"}' \
  | codegg exec --format json --quiet --approval-mode automatic --sandbox workspace-write

Sessions and diagnosis:

codegg sessions                     # list sessions
codegg export <session-id> -o s.json
codegg doctor search                # one subsystem
codegg auth status                  # stored credentials
codegg mcp list                     # configured MCP servers

In the TUI, space help lists every command (/models, /agent, /approval, /sandbox, /plan, /skills, /doctor, …). The full command and keybinding reference is in docs/tui.md.

Configuration

JSON with JSONC comments, discovered from $CODEGG_TUI_CONFIG, then system, global, and project locations (walked upward from the working directory):

{
  "model": "anthropic/claude-sonnet-4-20250514",
  "provider": {
    "anthropic": {
      "auth": { "type": "api_key", "env": "ANTHROPIC_API_KEY" }
    }
  }
}

Several config keys are untagged enums whose wrong shape silently discards your whole file — see docs/configuration.md.

Documentation

Guide Covers
docs/install.md Every installation path, upgrading, requirements.
docs/cli.md Every command, flag, and hidden compatibility spelling.
docs/configuration.md File locations, schema, model routing.
docs/providers.md Built-in providers, credentials, /connect.
docs/daemon.md The daemon, transports, scheduling.
docs/tui.md Sessions, slash commands, approval and sandbox.
docs/agents-skills.md Writing agents and SKILL.md skills.
docs/repository-init.md Read-only repository evidence and AGENTS.md draft semantics.
docs/tools.md Tool families, Git, deterministic tools, LSP.
docs/LSP.md Language-server configuration.
docs/MCP.md MCP servers, clients, transports.
docs/PLUGINS.md Plugin manifests, capabilities, SDK examples.
docs/themes.md Bundled themes and custom theme loading.
docs/playwright.md Playwright browser-testing plugins.
docs/security-semantics.md Command classification and escalation.
docs/TROUBLESHOOTING.md Symptom-driven fixes.
docs/execution-ownership.md Which component owns process spawning.
docs/dependency-maintenance.md Dependency and lint policy.
docs/windows-acl-leases.md Windows sandbox ACL lease semantics.
architecture/overview.md Architecture index and implementation contracts.
CHANGELOG.md Project changes.

Safety

CodeGG can read and modify files, run processes, operate Git, and contact configured services. Treat its permission configuration as part of your security boundary: /approval (interactive, automatic, yolo) and /sandbox (read-only, workspace-write, full-host) are visible in the status bar. Details in docs/security-semantics.md and docs/tui.md.

Development

scripts/verify.sh quick    # fmt, guard scripts, workspace check
scripts/verify.sh full     # the above plus clippy and the test suite
cargo fmt

Contributor conventions are in AGENTS.md.

License

MIT

About

AI harness for hatching new code

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages