Launch Claude Code against any LLM router.

What is leitum?

leitum is a small CLI tool that launches Claude Code against alternative LLM routers and providers (Requesty, OpenRouter, LiteLLM, local Ollama, etc.) instead of the Anthropic API directly. It mirrors the ergonomics of ollama launch and omlx launch: prepend leitum to your claude invocation and get your chosen provider and models configured automatically.

Install

# via uvx (no install needed)
uvx leitum --version

# via pip
pip install leitum

# via Homebrew
brew tap rgielen/taps
brew install leitum

Quickstart

# Initialize config directory and example providers file
leitum init

# Set your API key
export REQUESTY_API_KEY=your-key-here

# Launch Claude Code via Requesty
leitum claude

# Dry-run: see what would happen without launching
leitum --dry-run claude

Configuration

Config lives at ~/.config/leitum/api-providers.yaml (XDG). Example:

schema_version: 1
providers:
  - name: requesty
    base_url: https://router.requesty.ai
    auth:
      token: ${REQUESTY_API_KEY}
    defaults:
      start: anthropic/claude-sonnet-4-5
    models:
      - id: anthropic/claude-sonnet-4-5
        roles: [sonnet, start]
      - id: anthropic/claude-opus-4-5
        roles: [opus]
      - id: anthropic/claude-haiku-4-5
        roles: [haiku]

See docs/configuration.md for the full schema reference.

Pin provider and models per repository with leitum.yaml (checked into version control):

schema_version: 1
provider: requesty
models:
  start: anthropic/claude-sonnet-4-5

Or let leitum write it for you: run leitum -l claude, make your selection, and the resolved provider and models are saved to leitum.yaml (comments and extra_env in an existing file are preserved, and the global state is left untouched).

Project-local secrets with .leitumenv

Place a .leitumenv file in your project root to supply API tokens automatically on every leitum claude invocation — without direnv or manual shell exports.

Security: .leitumenv is executed as a bash script, so only use it in repositories you trust. Never check it into version control.

# .leitumenv — never commit this file

# Plain assignment (no shell needed)
export REQUESTY_API_KEY=rq-...

# Command substitution — fetch the token from 1Password on every launch
export REQUESTY_API_KEY="$(op read op://vault/requesty/Token)"

# Any bash expansion works
MY_CUSTOM_VAR=some_value

leitum sources this file via bash before loading any config, so ${VAR} references in api-providers.yaml and leitum.yaml resolve against its values. All bash expansions are supported, including $(...).

Behaviour:

• Shell environment takes precedence: a variable already set in your shell is never overwritten by .leitumenv.
• Lines starting with # and blank lines are ignored.
• A leading export prefix is accepted (but not required — set -a exports all assignments automatically).
• Use --no-dotenv to skip loading the file for a single invocation.

Add .leitumenv to your project's .gitignore to prevent accidental commits of tokens. leitum's own .gitignore already includes it.

CLI Reference

leitum [OPTIONS] <subcommand> [ARGS...]

Global options (before the subcommand):

Flag	Short	Effect
--provider <name>	-p	Set provider
--use-last-provider	-P	Reuse last provider
--model <id>	-m	Set START model
--use-last-model	-M	Reuse last START model
--opus <id>	-o	Set OPUS model
--sonnet <id>	-s	Set SONNET model
--haiku <id>	-k	Set HAIKU model
--refresh	-r	Refresh model cache
--no-project-config		Ignore leitum.yaml
--no-dotenv		Skip loading .leitumenv from CWD
--save-local	-l	Save the resolved selection to leitum.yaml instead of global state
--dry-run		Print env + exec line, do not launch
--verbose	-v	Verbose logging on stderr

During model selection, two keyboard shortcuts are available on every slot:

• Ctrl-R — re-fetch the provider's model list from the API and rebuild the choices.
• Ctrl-S — toggle saving the selection to leitum.yaml (equivalent to -l); the footer shows [save→project ON] when armed. Unavailable under --no-project-config.

Subcommands: claude, provider list/show/add/remove, refresh, doctor, init, completions.

See docs/commands.md for details.

Providers

• Requesty
• Local Providers (fully supported):
  • Ollama
  • LM Studio
  • llama.cpp, vLLM, and other local models

Troubleshooting

Run leitum doctor for a full sanity check. See docs/troubleshooting.md.

Contributing

See CLAUDE.md and the PRDs in prd/ for the authoritative specification. Also see the Roadmap for planned and shipped features.

License

Apache 2.0 — see LICENSE.