What contenox is

AI agents, under your command. contenox beam is the first-party terminal: a session at your own keyboard, with your filesystem and shell. contenox acp serves the same work over stdio to Zed, JetBrains, AionUi, OpenClaw and anything else that speaks the Agent Client Protocol. contenox run carries it into scripts, CI and unattended execution. No Contenox account is required.

Four things are in the box, and they are the product:

• The client. beam, in-tree, so the front door is never somebody else’s product. Bare contenox on a terminal opens it.
• The declaration. An agent is a Markdown file with a YAML frontmatter header, in .contenox/agents/. A directory of them is a workflow. The declarations you already keep for Claude Code, Copilot, Cursor, OpenCode or Antigravity are read where they are.
• The compile step. Each declaration becomes a chain (what happens) and an envelope (what is permitted) — JSON, schema-validated, yours to read, neither yours to maintain.
• The tool boundary. contenox ships no tools. Your editor or terminal client supplies fs/* and terminal/*; you attach MCP servers and OpenAPI services; every call crosses one boundary before it runs.

SQLite provides the zero-dependency default for state. Deployments that need operated infrastructure can use PostgreSQL for the store, NATS for messaging and Valkey for key-value state. Native inference runs through modeld; Ollama, vLLM and hosted providers can be used alongside it. Telemetry remains opt-in.

Why I built it

I wanted to hand an agent a job and walk away — and the thing that stopped me was never whether the model was good enough. It was that I had no way to say this directory is untouchable and know it would hold when the model decided a cleanup was in order.

So the answer had to be a file I wrote, that I could read back, and that was enforced by something other than the model’s good intentions. Once that file existed the rest followed: the same file names what may not proceed without me, and the things that start work are declared rather than discovered.

There is a second reason, and it is not smaller. I did not want the record of how my work behaves to live in someone else’s dashboard — a product that can pivot, exit, leak, or turn my usage into their next feature. It runs on my machine. That is a choice about who holds the evidence, not a deployment preference.

What it is typically used for

• Working in a repository — review a diff, plan a change, make it, run the tests, fix what broke, all in beam or in the editor you already use.
• Repeatable jobs — a review pass, a release-notes draft, a dependency sweep — declared once and fired with contenox run from a shell, CI or cron.
• Reaching systems the model should not hold keys for — an internal API wrapped as a tool with its sensitive arguments filled in by config.
• Work that must stop for a person — anything that sends, deploys or deletes, held at the tool boundary until someone answers.
• Local inference where nothing may leave the network, or a hosted model on your own key and your own region — a config change, not a rewrite.

What it is not

• Not a chat product. There is no thread to keep warm and nothing that wants you back tomorrow.
• Not a dashboard. No screen summarises what an agent did; if you want to know, you read the captured execution state on your own disk.
• Not a hosted service. No Contenox service sits in the execution path. External calls go only to the model backends and tools you configured.
• Not an editor plugin. ACP is a protocol the editor speaks; the terminal client needs no editor at all.
• Not a framework or an SDK. You run it; you do not import it.
• Not autonomous. It does exactly what you declared and stops where you said stop. That is the feature.

Where to go next

• Declaring agents — the file, the frontmatter, and what agents.toml supplies that a declaration cannot.
• Human-in-the-loop policies — the envelope in full.
• How contenox compares — API integration, workflows, and execution policy.
• Envelope JSON Schema and chain JSON Schema — the formats, generated from the code that loads them.
• Sovereignty — local inference, EU regions, and what the EU AI Act asks of whoever deploys.