altcode — ~/project
The open agent harness for coding
One binary · Any model · 13 providers · Production-grade agent infrastructure
────────────────────────────────────────────────────────────────

Install

curl -fsSL https://altcode.io/install.sh | bash

Also: brew install jiayaoqijia/tap/altcode · go install github.com/altcode-ai/altcode/cmd/altcode@latest

Binaries: Linux · macOS · Windows — no runtime deps, single binary

────────────────────────────────────────────────────────────────

Zero Config

Already use Claude Code or Codex CLI? Just run altcode — credentials auto-detected.

If you have...altcode uses it
Claude Code CLIClaude subscription ✓
Codex CLICodex subscription ✓
ANTHROPIC_API_KEYAnthropic API ✓
OPENAI_API_KEYOpenAI API ✓
DEEPSEEK_API_KEYDeepSeek ✓
ZHIPU_API_KEYGLM-5 ✓
MOONSHOT_API_KEYKimi K2.5 ✓
MINIMAX_API_KEYMiniMax M2.7 ✓
DASHSCOPE_API_KEYQwen ✓
ALTLLMaltllm-basic ✓
OpenRouter key100+ models ✓
Ollama runningLocal models ✓
────────────────────────────────────────────────────────────────

Usage

altcode
# Interactive TUI (auto-detects provider)
altcode "explain this error"
# Headless exec mode
altcode --model deepseek/deepseek-chat
# Chinese providers: DeepSeek, GLM, Kimi, MiniMax, Qwen
altcode --model ollama/llama3
# Local models, no API key needed
altcode workflow --mode ralph "add auth"
# Persistent execution — loops until complete
altcode workflow --mode interview "add rate limiting"
# Socratic clarification before coding
────────────────────────────────────────────────────────────────

Why a harness?

Coding CLIaltcode (Harness)
Agent loopRun onceMulti-turn + verification gates
ContextSend & hopeToken tracking + auto-compact + LLM summarization
ToolsCall & prayPermissions + hooks + auto-verify
AgentsSingle modelMulti-agent + mailbox IPC + roles
WorkflowAd-hocInterview → plan → persistent execution
Providers1 vendor13 providers, any model
────────────────────────────────────────────────────────────────

Benchmarks (5 suites × 6 models)

ModelScore
DeepSeek V396%
MiniMax 2.793%
Qwen Coder93%
Claude Haiku90%
GLM-576%
Kimi K2.540%

HumanEval + SWE-bench + Terminal-Bench + Aider + FeatureBench · All via altcode

────────────────────────────────────────────────────────────────

Features

Multi-turn agent loop (tool dispatch, 50-iter cap)
7 tools: read, glob, grep, ls, bash, edit, write
MCP client (stdio + SSE) · 13 hook events
Slash commands · Plugin system · Marketplace
Context harness — auto-compact, LLM summarization, pre-turn compaction
Agent harness — multi-agent, mailbox IPC, history forking, roles
Workflow harness — interview → plan → persistent execution (ralph)
Tool harness — permissions, hooks (command + HTTP), auto-verify
47 skills · OAuth login · Persistent memory · Claude Code compat

────────────────────────────────────────────────────────────────

Links

GitHub · Releases · Docs · License

altcode
● ready
Enter send · Esc quit
v0.6.0