oh-my-code

A Rust-native interactive terminal coding assistant. Rebuilds the Claude Code experience in Rust with a pluggable provider layer, a native search toolchain, and a concurrent tool executor with read/write safety partitioning.

Features

• Multi-provider: Claude (Anthropic Messages API), OpenAI / GPT, Zhipu, MiniMax (OpenAI-compatible endpoint), and MiniMax (Anthropic-compatible endpoint at api.minimaxi.com/anthropic with Bearer auth). Add a new provider by extending create_provider in src/model/mod.rs.
• Two configuration paths: a three-variable .env quick start (API_KEY + BASE_URL + MODEL, with wire format and auth style auto-detected from the URL) that coexists with a full config.toml multi-provider setup for users who want to juggle several backends.
• Streaming agent loop: Assistant text streams to the terminal as it arrives; tool calls are dispatched and fed back into the conversation until the model stops calling tools.
• 11 built-in tools: think, grep, glob, file_read, file_edit, file_write, bash, enter_plan_mode, exit_plan_mode, web_fetch, web_search.
• Concurrent tool execution with safety partitioning: Read-only tools run in parallel via join_all; write tools run sequentially in arrival order.
• Plan / Act modes: An atomic shared flag gates writes. In Plan mode, the model can read and reason but cannot mutate files or run shell commands.
• Native search toolchain: Uses grep-searcher + grep-regex + ignore + syntect directly — no shelling out to rg or fd.
• Interactive REPL: rustyline-based line editor with slash commands (/help, /model, /session, /clear, /quit).
• File-based sessions: Conversations are persisted as JSON under ~/.config/oh-my-code/sessions/.

Install

Requires a recent stable Rust toolchain (install via rustup).

git clone <this-repo>
cd oh-my-code
cargo build --release

The binary lands at target/release/oh-my-code.

Configuration

There are two ways to configure oh-my-code. They coexist — the quick-start env vars win when all three are set; otherwise config.toml drives provider selection.

Quick start (.env, three variables)

cp .env.example .env
# then edit .env and fill in:
#   API_KEY=sk-...
#   BASE_URL=https://api.anthropic.com          # or any OpenAI / Anthropic-compatible endpoint
#   MODEL=claude-sonnet-4-5

oh-my-code reads .env at startup via dotenvy, then auto-detects the wire format and auth header from BASE_URL:

If any of the three vars is missing or empty, the quick-start path stays inactive and config.toml is used instead. Partial configuration (e.g. API_KEY and BASE_URL set but MODEL missing) prints a warning to stderr so you know why your overrides are not taking effect.

Multi-provider via config.toml

On first run, a default config is written to ~/.config/oh-my-code/config.toml. Edit it to change the active provider, model, search ignore patterns, or session storage directory. The in-repo template lives at config/default.toml. Each provider reads its API key from the env var named in its api_key_env field:

Usage

# Option 1 — quick start (reads .env)
./target/release/oh-my-code

# Option 2 — inline env var, provider chosen in config.toml
ANTHROPIC_API_KEY=<key> ./target/release/oh-my-code

Once in the REPL, type a request in natural language. Useful slash commands:

• /help — list commands
• /model — switch model
• /session — list / load / save sessions
• /clear — clear the current conversation history
• /quit — exit

Development

cargo test              # full test suite
cargo test <path>       # run a subset, e.g. cargo test model::types
cargo clippy --all-targets
cargo fmt

Note: oh-my-code is a binary-only crate, so cargo test --lib ... will fail — use cargo test <module_path> instead.

See CLAUDE.md for a deeper architectural tour.

License

Licensed under the Apache License, Version 2.0. See LICENSE for the full text.