contenox

Contenox

AI workflows at your fingertips

Contenox is a lightning-fast, fully-local CLI that turns natural language goals into persistent, step-by-step plans and executes them with real shell + custom hooks like filesystem tools. Powered by any LLM (Ollama, OpenAI, Gemini, vLLM, etc.). Zero cloud required.

$ contenox plan new "install a git pre-commit hook that prevents commits when go build fails"
Creating plan "install-a-git-pre-commit-a3f9e12b" with 5 steps. Now active.

$ contenox plan next --auto
Executing Step 1: Install necessary tools...              ✓
Executing Step 2: Create .git/hooks/pre-commit...         ✓
Executing Step 3: Edit the hook script with the check...  ✓
Executing Step 4: Write bash content to the hook file...  ✓
Executing Step 5: chmod +x .git/hooks/pre-commit...       ✓

No pending steps. Plan is complete!

The model wrote that hook. On your machine. No copy-paste hell.

⭐ Leave us a star if you like it! | 🌟 We welcome any suggestions, and contributions!

📺 contenox vibe — Interactive TUI

When you want more than a shell prompt:

contenox vibe

A full-screen terminal dashboard (Bubble Tea) with:

• Live plan sidebar — watch steps execute with ⟳ / ✓ / ✗ indicators in real time
• Interactive approvals — approve or deny sensitive filesystem actions with y/n before they run
• Full CLI parity — every contenox subcommand is a slash command inside vibe: /plan, /model, /session, /backend, /hook, /mcp, /config, /run

/plan new "add prometheus metrics to the HTTP server"
/plan next --auto              ← run to completion
/model set-context gpt-5-mini --context 128k
/backend add local --type ollama --url http://127.0.0.1:11434
/mcp add memory --transport stdio --command npx --args "-y,@modelcontextprotocol/server-memory"
/help

Why Contenox?

Contenox is different:

• Persistent plans stored in SQLite — pause, inspect, retry, replan at any time
• Human-in-the-loop by default — --auto only when you say so
• Real tools — shell commands and filesystem, not just code suggestions
• Fully offline with Ollama — no data leaves your machine
• Chains are just JSON — write your own LLM workflows
• Workflow Engine — Contenox is not a toy, a complete statemachine lives under the hood.
• Native MCP Support — connect to local filesystems, memory servers, and remote tools instantly via the Model Context Protocol.

🔌 Universal Tooling with MCP

Contenox is a native Model Context Protocol (MCP) client. Instead of writing custom integrations, you can instantly connect your local agent to any MCP-compatible data source, persistent memory, or tool API.

# Give your agent access to the local filesystem
contenox mcp add filesystem --transport stdio \
  --command npx --args "-y,@modelcontextprotocol/server-filesystem,/"

# Give your agent a persistent memory graph across reboots
contenox mcp add memory --transport stdio \
  --command npx --args "-y,@modelcontextprotocol/server-memory"

# Connect to cloud tools securely over SSE
contenox mcp add cloud-tools --transport sse --url https://api.example.com/mcp

Every registered MCP server becomes natively available to your agent during chat sessions and execution plans.

🛠 Turn Any API into an Agent Tool

Don't need the MCP ecosystem? Expose any HTTP API as an agent tool in seconds with contenox hook add. Write a FastAPI service — Contenox reads its OpenAPI schema and makes every endpoint callable by the model, with no extra glue code.

# Register your FastAPI service as a tool
contenox hook add my-api --url http://localhost:8000

# The model can now call any endpoint on it directly as a tool
contenox run "fetch the latest metrics from my API and summarize them"

Any service that speaks HTTP and exposes an OpenAPI spec becomes a first-class agent tool.

Quick Start

Install

Ubuntu / Linux

TAG=v0.6.1
ARCH=$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/')
curl -sL "https://github.com/contenox/contenox/releases/download/${TAG}/contenox-${TAG}-linux-${ARCH}" -o contenox
chmod +x contenox && sudo mv contenox /usr/local/bin/contenox
contenox --version

macOS

TAG=v0.6.1
ARCH=$(uname -m | sed 's/x86_64/amd64/;s/arm64/arm64/')
curl -sL "https://github.com/contenox/contenox/releases/download/${TAG}/contenox-${TAG}-darwin-${ARCH}" -o contenox
chmod +x contenox && sudo mv contenox /usr/local/bin/contenox
contenox --version

Or pick a binary from Releases.

First run

# 1. Initialize (creates .contenox/ with default chains)
contenox init

# 2. Register a backend
ollama serve && ollama pull qwen2.5:7b
contenox backend add local --type ollama
contenox config set default-model qwen2.5:7b

# Or for OpenAI / Gemini:
# contenox backend add openai --type openai --api-key-env OPENAI_API_KEY
# contenox config set default-model gpt-5-mini

# 3. Chat with your model:
contenox "hey, what can you do?"
echo 'fix the typos in README.md' | contenox

# 4. Plan and execute a multi-step task:
contenox plan new "create a TODOS.md from all TODO comments in the codebase"
contenox plan next --auto

Requirements: an LLM with tool calling support. Local: ollama serve && ollama pull qwen2.5:7b Cloud: register a backend with contenox backend add and set your API key via --api-key-env.

Full example

# 1. Create
contenox plan new "install a git pre-commit hook that blocks commits when go build fails"

# 2. Review the plan before touching anything
contenox plan show

# 3. Execute one step at a time
contenox plan next
contenox plan next
# ...

# Or run everything at once once you trust it
contenox plan next --auto

# 4. If a step went wrong
contenox plan retry 3

# 5. Final check
contenox plan show

contenox plan — AI-driven plans

contenox plan new "migrate all TODO comments in the codebase to TODOS.md"
contenox plan new "set up a git pre-commit hook that blocks commits when go build fails"
contenox plan new "find all .go files larger than 500 lines and write a refactoring report"

Contenox breaks any goal into an ordered plan, then executes it step by step using real tools.

Commands

Pro tip: Always do contenox plan show before --auto.

contenox chat — Persistent chat session

contenox chat "what is my current working directory?"
contenox chat "list files in my home directory"
echo "explain this" | contenox chat

Uses .contenox/default-chain.json. Natural language → shell tools → response.

contenox run — Scriptable, stateless execution

For CI/pipelines where you want full control:

contenox run --chain .contenox/my-chain.json "what is 2+2?"
cat diff.txt | contenox run --chain .contenox/review.json --input-type chat
contenox run --chain .contenox/doc-chain.json --input @main.go
contenox run --chain .contenox/parse-chain.json --input-type json '{"key":"value"}'

run is stateless — no chat history. --chain is required. Supported --input-type: string (default), chat, json, int, float, bool.

🧠 Reasoning model support

Pass --think to stream the model's internal chain-of-thought to stderr before it acts — works with DeepSeek-R1, OpenAI o3, Gemini Thinking, and Ollama thinking models:

contenox --think "why is my API slow?"
contenox run --chain .contenox/review.json --think --input @main.go

Configuration

Contenox stores all configuration in SQLite (.contenox/local.db or ~/.contenox/local.db). No YAML file needed — register backends and set defaults using CLI commands.

Register a backend

# Local Ollama (URL inferred automatically)
contenox backend add local --type ollama

# OpenAI (base URL inferred)
contenox backend add openai --type openai --api-key-env OPENAI_API_KEY

# Google Gemini
contenox backend add gemini --type gemini --api-key-env GEMINI_API_KEY

# Self-hosted vLLM or compatible endpoint
contenox backend add myvllm --type vllm --url http://gpu-host:8000

Set persistent defaults

contenox config set default-model    qwen2.5:7b
contenox config set default-provider ollama
contenox config set default-chain    .contenox/default-chain.json

contenox config list   # review current settings

Manage backends

contenox backend list
contenox backend show openai
contenox backend remove myvllm

Safety

• Opt-in shell access — --shell flag must be passed explicitly to enable local_shell
• Chain-scoped policy — allowed and denied commands are declared in the chain's hook_policies field; the default chains ship with a sensible allowlist out of the box
• Human-in-the-loop — plan next executes one step and stops; --auto requires explicit intent
• Local-first — with Ollama, nothing leaves your machine

Architecture

contenox CLI
  ├── plan new       → LLM planner chain → SQLite plan + steps
  ├── plan next      → LLM executor chain → local_shell / local_fs → result persisted
  ├── vibe           → Bubble Tea TUI: chat + live plan sidebar + HITL approvals
  ├── run            → run any chain, any input type, stateless
  ├── (bare)         → stateless run via default-run-chain.json (same as run)
  └── chat           → LLM chat chain → session history persisted in SQLite

SQLite (.contenox/local.db)
  ├── plans + plan_steps   (autonomous plan state)
  ├── message_index        (chat sessions)
  └── kv                   (active session + config)

Chains are JSON files in .contenox/. They define the LLM workflow: model, hooks, branching logic. See ARCHITECTURE.md for the full picture.

Contenox is powered by a battle-tested enterprise workflow engine. The Runtime API is also available as a self-hostable Docker deployment for teams who want the full server with REST API, observability, and multi-tenant support.

Build from source

git clone https://github.com/contenox/contenox
cd contenox
go build -o contenox ./cmd/contenox
contenox init

Questions or feedback: hello@contenox.com