中文

Claude Code on the ChatGPT plan you already pay for.

alc is one small CLI that launches nine coding agents on a single login, or on any of sixteen providers you point them at. No API key. No config file.

curl -fsSL https://raw.githubusercontent.com/treeleaves30760/all-code/main/install.sh | sh
Read the docsSource on GitHub

Version 2.1.0. MIT licensed. Rust, one binary.

Plug any agent into any provider.

Agents
Providers
Logins you have
Hosted APIs
On your machine
alc --codex claude

Claude Code reaches Codex over Anthropic Messages. Through alc’s built-in Codex bridge on a loopback port. No key leaves your machine.

  • Anthropic Messages
  • OpenAI Responses
  • Chat Completions
  • native login

Three commands. Nothing to configure.

The starter configuration is compiled into the binary and already carries a Codex profile. alc reads it in memory and writes no config.toml.

  1. Install alc

    curl -fsSL https://raw.githubusercontent.com/treeleaves30760/all-code/main/install.sh | sh

    The installer verifies the SHA-256 of the release and puts alc in ~/.local/bin.

  2. Sign in to Codex

    codex login

    Uses the ChatGPT plan you already have. alc reads the auth.json it writes and never copies it.

  3. Launch Claude Code

    alc --codex claude

    It opens on gpt-6.1-sol at low effort with the model’s real 500K context window.

Every Codex model in Claude Code’s own /model picker

/modelSelect model

gpt-6.1-solLatest workhorse; the recommended start500K
gpt-6-astraComplex, demanding work272K
gpt-6-solEveryday coding and agentic work272K
gpt-6-lunaFast and cheapest; quick fixes272K
gpt-5.6-solPrevious generation, professional work272K
gpt-5.6-terraPrevious generation, balanced272K
gpt-5.6-lunaPrevious generation, fast272K

Left and right arrows move the effort slider. --context max raises any model to an 872K window.

One login, every agent.

alc starts its own Codex adapter on a loopback port and points only the launched agent at it. Three wire protocols, one codex login, no second key.

AgentCodex bridge on 127.0.0.1Reaches the bridge overSwitching models
Claude Codealc claudeAnthropic Messages/model picker, mid-session
OpenCodealc opencodeOpenAI ResponsesOne model, chosen at launch
Pialc piOpenAI ResponsesOne model, chosen at launch
Kimi Code CLIalc kimiOpenAI ResponsesOne model, chosen at launch
Copilot CLIalc copilotChat CompletionsOne model, chosen at launch
Goosealc gooseChat CompletionsOne model, chosen at launch
Qwen Codealc qwenChat CompletionsOne model, chosen at launch
Codex CLIalc codexnative loginCodex’s own picker, no adapter
Antigravity CLIalc agynot bridgedIts own Google login

Drive it from your phone.

Any session, any agent, any provider, mirrored to a browser page. Your terminal keeps working: sharing mirrors a session, it does not take it away.

  • Keys a phone keyboard lacks

    Esc, Tab, Shift+Tab, Ctrl and arrows sit on a key bar, and a composer sends a whole prompt as one block.

  • Sessions outlive the terminal

    A background hub owns them. alc sessions lists them, alc attach 7QK2 puts any terminal back on one.

  • The link is the credential

    The token lives in the URL fragment, so it never reaches a server, proxy or access log. Treat it like a password.

Permission rungs, strictest first

A shared Claude Code, Codex, OpenCode or agy session starts at ask. The page can loosen it up to a ceiling; past that, the change waits for alc confirm typed at a terminal on the host, which an agent cannot do for itself.

  1. planReads and plans, writes nothing
  2. askAsks before anything that changes the worldShared sessions start here
  3. auto-editEdits files, still asks for commandsDefault ceiling
  4. autoActs inside the agent’s sandbox
  5. fullNo gate

alc confirm 7QK2M9XB4T

Reach it from anywhere

alc --share --bind-lan claude

Nothing to install. Plain HTTP, so keep it to a network you trust.

See what is left before you run out.

alc usage asks each login’s own vendor what remains, reads nothing back and refreshes no token, so a status check can never break a running session.

$ alc usage

Accounts

anthropic~/.claudemaxweek 79% left, resets in 6d 4h
codexyou@example.comproweek 66% left, resets in 5d 9h
openrouterworkcredits$20.50 of $50 left
ollamalocalhost—no quota API
alc usage weekly --chart
Daily tokens and estimated cost across Claude Code, Codex and every alc launch, priced from a pinned, offline LiteLLM snapshot.
alc tps
Time to first token and tokens per second for your latest requests, measured where alc carries the traffic.
alc usage yearly --wrapped
One shareable PNG of the year: busiest day, streaks, top models, cache hit rate and peak hour.

Local models that do not time out.

Claude Code opens every session with roughly 25k to 40k tokens of system prompt and tool schemas. A laptop-sized model reads that at a few dozen tokens per second.

Time to first token, gemma4:12b on an M3 MacBook Air
22k-token request6 min
39k-token request15 min
Claude Code gives up and retriesalc’s first-token timeout

alc --ollama claude, alc --llamacpp claude and alc --vllm claude pin every model alias to the one model the server loaded, turn off non-essential traffic, read the real context window from the server and raise the first-token timeouts to thirty minutes.

alc doctor checks the server, whether the model is pulled, whether it can call tools, and the context it actually gets.

alc --ollama claude
alc --llamacpp claude
alc --vllm claude

Plain about what it is.

  • A third-party compatibility layer, not an official OpenAI, Anthropic or Google integration. Read THIRD_PARTY.md and your provider’s terms before routing subscription credentials through it.
  • Credentials stay where your tools put them. alc reads ~/.codex/auth.json and never copies it into its own config.
  • alc reads no configuration from the working repository, so a checked-in file can never turn sharing on.
  • alc prints nothing at launch. What you are looking at is the agent itself.

Questions

Do I need an API key to run Claude Code with alc?

Not on the Codex path. alc needs the auth.json that codex login writes and the claude binary on your PATH. It translates Claude Code’s Anthropic Messages requests to your ChatGPT plan through a bridge compiled into alc itself.

Does my normal claude command still use Anthropic afterwards?

Yes. alc records the model key in ~/.claude/settings.json before launch and restores it when the session exits, so sessions alc did not start are untouched.

Which coding agents does alc launch?

Nine: Claude Code, Codex CLI, OpenCode, Pi, Copilot CLI, Goose, Qwen Code, Kimi Code CLI and Antigravity CLI. alc launches agents you have installed; it does not bundle them.

Which providers can I point them at?

Sixteen provider kinds: Codex, Antigravity, Anthropic, OpenAI, OpenRouter, Ollama, vLLM, llama.cpp, DeepSeek, Moonshot, Z.ai, MiniMax, Groq, xAI, Google and any custom endpoint. alc checks each agent and provider pairing before launch.

Is alc an official OpenAI or Anthropic tool?

No. It is an independent, MIT-licensed open-source project. The bridge is a third-party compatibility layer, so review your provider’s terms before routing subscription credentials through it.

Does it work on Windows?

Yes, on Windows 10 and 11 with PowerShell 5.1 or 7, as well as macOS, Linux and WSL. Remote control and --tmux sessions work on Windows too, using the native Windows port of tmux.

Can I run Claude Code on a local model?

Yes, through Ollama, llama.cpp or vLLM. alc pins the model aliases to the loaded model, reads the real context window and raises the first-token timeouts to thirty minutes so a slow first prompt is not abandoned.

How much does alc cost?

Nothing. alc is free and MIT licensed. You keep paying only for the subscriptions and API usage you already have.

Plug in.

One install, one login, nine agents.

curl -fsSL https://raw.githubusercontent.com/treeleaves30760/all-code/main/install.sh | sh
ReleasesSource on GitHub