AI coding agent CLI by Huanime. 22 LLM providers, 101 models, MIT-licensed.

A production-grade AI coding agent in your terminal. Stream-native engine, modern TUI, type-safe multi-provider abstraction. Built for developers who want Claude-Code-class capabilities with full provider flexibility.

huagent ·                                                                 claude-sonnet-4-6
 autonomous off  scope src/middleware/jwt.ts  perm workspace-write
──────────────────────────────────────────────────────────────────────────

> add jwt auth to our express app, replace the old session thing
✓ Got it. Let me first scope the existing auth, then design a JWT middleware

✓ PLAN    add jwt auth to our express app          1.0s
✓ READ    src/middleware/auth.ts                    0.8s
✓ WRITE   src/middleware/jwt.ts                     2.0s
✓ TEST    src/middleware/jwt.ts: passed             2.3s

╭─ subagents (1 running) ─╮  ╭─ Tests passed: 12/12 in 2.3s ─╮
│ reviewer   Check ... 65%│  ╰───────────────────────────────╯
╰─────────────────────────╯

╭─ v4 │ tok 4.2k ($0.0127) │ steps 8 │ perm workspace-write ─╮
╰─────────────────────────────────────────────────────────────╯

One-liner (recommended):

curl -fsSL https://raw.githubusercontent.com/ahmdd4vd/Huagent/main/install.sh | sh

npm:

npm install -g huagent

From source:

git clone https://github.com/ahmdd4vd/Huagent.git
cd Huagent
npm install
npm run build
node bin/huagent.js

export ANTHROPIC_API_KEY=***      # or any other provider
huagent

huagent auto-detects your provider from environment variables:

22 LLM providers, 101 models. Each provider ships with a default model, full model registry, and capability metadata.

Flagship: Claude Opus 4.7, Claude Sonnet 4.6, GPT-5.5, Gemini 3 Pro, Grok 4, Llama 3.3 70B (via Groq/Cerebras/Together/Fireworks/OpenRouter)

Reasoning: DeepSeek R1, o3, QwQ 32B, Sonar Reasoning, Magistral Medium

Fast: Claude Haiku 4.5, GPT-5 mini, Llama 3.1 8B Instant, Gemini 3 Flash, Grok 3 mini

Code: Qwen 3 Coder 480B, Codestral 25, DeepSeek Coder V2, GPT-5.5 Codex

Local (free): Ollama (Llama 3.2, Qwen 2.5 Coder 32B, DeepSeek R1 32B)

Search-augmented: Perplexity Sonar Pro

Cloud: AWS Bedrock (Claude 4.7, Nova Pro/Lite, Llama 3.3 70B), Google Vertex AI (Claude 4.6, Gemini 3 Pro)

Browse all in the TUI:

huagent             # then type:
/providers          # list all 22 providers
/models             # models for current provider
/models anthropic   # models for a specific provider
/models openai gpt-5.5  # switch model

• 22 LLM providers out of the box, 101 models with pricing + capabilities
• Stream-native engine — type-safe SSE, automatic tool-call accumulation, accurate cost tracking
• Modern TUI — width-adaptive (40–240+ cols), no emoji, modern palette
• Discipline layer — every task goes through Plan → Ground → Observe → Diagnose → Verify
• WllmConcept — wiki knowledge engine with semantic search, evolution, provenance
• 26 slash commands — full runtime control over model, provider, mode, scope, memory
• 5 permission modes — read-only, workspace-write, sandboxed, danger-full-access, custom
• Auto-detect — set an env var, huagent figures out the provider
• MIT licensed — free forever

huagent                                            # interactive TUI
huagent "fix the auth bug"                         # one-shot, no TUI
huagent --provider anthropic --model claude-sonnet-4-6
huagent --autonomous                               # start in autonomous mode
huagent --scope src/auth.ts                        # limit to one file
huagent --permission-mode sandboxed
huagent --tui=modern                               # default TUI
huagent --tui=legacy                               # original TUI
huagent --version

870+ tests across 5 suites:

npm test

npm run verify     # lint + test + build

See docs/ARCHITECTURE.md for a high-level map of the codebase.

src/
├── cli.tsx              # entry point, arg parsing, bootstrap
├── providers/           # 22-provider registry, 101-model list, unified client
├── engine/v4/           # stream-native actor model, discipline layer
├── tui/                 # modern TUI (width-adaptive)
├── tools/               # bash, file ops, git, search
├── memory/              # SQLite-backed memory
├── wllm/                # wiki concept engine
├── sessions.ts          # session save/load
└── slash-commands.ts    # 26 slash commands

See CONTRIBUTING.md. Bug reports and feature requests welcome via issues.

See SECURITY.md for the vulnerability reporting policy and threat model.

MIT © 2026 Huanime. All rights reserved.

Inspired by claw-code, OpenClaude, and ECC. Built with Ink.