▍ fig.01 · Independent AI research lab
Zerolinear

An AI research lab building open agentic systems.

We research how autonomous agents reason, use tools, and stay under human control — and we ship what we learn as open-source software. Our first system is Zeline, with more products to come.

Focus
Agentic AI
First system
Zeline
License
Open source
Approach
Model-agnostic
I — About the lab

A research lab, not an app studio.

We study the systems behind autonomous agents — reasoning, memory, tool-use, and control — and turn that research into open, reusable software.

Research first

We start from open questions about how agents should think and act, then build to test them.

Building Zeline

Our first system, Zeline, turns that research into something developers can run.

More ahead

Zeline is the foundation. New tools grow on top as the lab expands.

First system — Framework

Zeline — an open-source agentic framework.

A foundation for agents that reason, use tools, and run multi-step workflows — while you keep control of models, tools, and data.

Agent core

OpenAI-compatible agent loop with tool calling and an interactive CLI.

Model-agnostic

OpenAI, OpenRouter, vLLM, Ollama — swap without rebuilding.

Persistent memory

Long-term memory per identity, stored locally, surviving restarts.

II — Skills, tools & integrations

What Zeline can do.

01 — SKILLS

Reusable skills

Markdown procedures loaded on demand — new workflows, no code.

02 — TOOLS

Built-in tools

Web search, deep research, file I/O, code execution, and shell.

03 — MCP

MCP client

Connect external MCP servers and expose their tools automatically.

04 — GATEWAYS

Messaging gateways

Telegram, WhatsApp, and a local webhook — one agent, many surfaces.

05 — PROFILES

Scoped profiles

safe · workspace · full — gate host and shell access per surface.

06 — SESSIONS

Session persistence

History in SQLite, so context survives restarts.

TELEGRAMWHATSAPPWEBHOOKCLI MCP CLIENTOPENAIANTHROPICOLLAMAvLLM
III — Vision & mission

Keep control in the developer's hands.

Agentic AI should amplify people, not lock them in. Our mission: make autonomous systems open, portable, and controllable — your models, your tools, your data.

Open & inspectable

No black boxes. Every layer is open-source and self-hostable.

Portable by design

Models and providers swap freely — the architecture outlives any vendor.

Human-in-control

Scoped permissions keep people in charge of what agents can do.

IV — How it's built

Built in Python, runs anywhere.

A lightweight framework with no heavy dependencies — run it locally or deploy to your own server.

01

Connect a model

Point Zeline at any OpenAI- or Anthropic-compatible endpoint.

02

Choose a surface

Run the CLI, or set up a Telegram / WhatsApp / webhook gateway.

03

Extend it

Add built-in tools, MCP servers, and Markdown skills.

04

Deploy & keep control

Ship it anywhere — scoped profiles keep access safe.

V — Roadmap

What the lab is building next.

NOW

Zeline framework

Maturing the agent core, gateways, and tool ecosystem.

NEXT

Autonomous jobs

Long-running, self-directed workflows and richer plugin APIs.

LATER

New products

Purpose-built tools layered on the framework as the lab grows.

VI — Get started

Everything about Zeline lives in the docs.

Installation, providers, gateways, skills, tools, and API reference — all in one place.