RELENTLESSWORKS

WHY RELENTLESSWORKS

relentlessworks is an experiment in autonomous software development. The goal: a software company that designs, builds, ships, and maintains its own products — with AI agents doing the work end to end.

No human writes the code. No human files the tickets. An agent gets an idea, opens a repo, writes the service, sets up CI, deploys it, and iterates. The entire lifecycle runs on its own.

To make that possible, every product is built agentic-first — designed for AI agents to drive over plain HTTP. No UI, no SDK, no dashboards. Plain-text responses that are grepable and token-cheap. Self-documenting with a one-page manual at /help. Instructive errors that tell the agent what to do next.

Each service is a single static Go binary with zero external dependencies. Deploy as one file. Zero config defaults. Multi-tenant ready with workspaces, plans, limits, and audit logs built in.

The agent is the developer. The agent is the user. The agent is the interface.

HOW IT WORKS

Everything the agent does is fully recorded and publicly inspectable. Every decision, every line of code, every commit — nothing happens behind closed doors.

The agent's full session history is available on the ChatBotKit blueprint page. You can watch it work in real time: read its thought process, see the commands it runs, review the code it writes, and trace every change from idea to deployment.

The source code for every service lives on GitHub, with full commit history and CI logs. The blueprint page shows the agent side — the reasoning, the iterations, the mistakes and fixes.

View the agent's full history

https://chatbotkit.com/hub/blueprints/coder

AGENTIC-FIRST SERVICES

Services for AI.
Not for Humans._

relentlessworks is an autonomous software company. AI agents design, build, ship, and maintain every product — end to end, no humans in the loop. Each service is a single Go binary with a plain-text, grepable API designed for agents to drive over plain HTTP. No dashboards, no clicks — the agent is the interface.

// principles

The agent IS the interface

No UI, no SDK. The AI agent is the user. The API is the product.

Plain text by default

One labeled, grepable line per record. Token-cheap. JSON on demand.

Instructive errors

Every 4xx includes a hint telling the agent what to do next.

Self-documenting

GET /help returns a one-page operating manual the agent loads at runtime.

Simple auth

OTP via email → long-lived bearer token. No complex OAuth flows.

Single static binary

Go + SQLite, zero external dependencies. Deploys as one file.

Zero config defaults

Runs out of the box. Config layered: defaults < file < env < flags.

Multi-tenant ready

Workspaces, plans, limits, and audit logs built in from the start.

MCP connector

Also speaks Model Context Protocol at /mcp for chat client integrations.

Short stable handles

Every record is addressed by a short workspace-scoped handle like contact_k7m2q.

// how it works

1. Point your agent at a service
Give it the URL or ask it to read /help

2. The agent reads the manual
A one-page operating manual at /help tells it everything

3. OTP login → bearer token
The agent authenticates via email and saves the token

4. Just talk to it
The agent drives the API over plain HTTP

Then just talk to it:

> Add a new contact from ACME Corp

> Show me all tasks due this week

> Create a short link for this URL

> Search my notes for "deployment strategy"