About

Blondie is an autonomous coding agent.

It reads tasks from a backlog, claims one, executes it via LLM-driven tool calls, then commits and merges - and picks the next.

What it does

Blondie attaches to a repository and runs the engineering loop: plan, design, implement, review, test, verify, document, integrate, learn. Each step is a skill the agent invokes; skills can call other skills, so a task as small as a typo fix and as large as a feature ship use the same machinery.

How it works

Tasks live in a pluggable backlog - the local TASKS.md, GitHub Issues, Jira, or a custom backend. The orchestrator picks a task, prepares a workspace, and hands it to a reasoning loop that uses a small, audited toolset to read code, write code, run checks, and commit.

Every action is gated by policy. The agent cannot bypass hooks, force-push, or perform a destructive operation without an explicit authorisation it can audit. The same mechanism lets you grant looser autonomy in trusted scopes and tighter autonomy at the edges.

Why an agent

Most engineering work isn't novel - it's the assembly of known patterns against a moving target. An agent that can hold the whole picture, take a task end-to-end, and learn from each pass is the right shape for the next decade of software work. Blondie is a bet on that shape.

Get in touch

More to come. Pricing, docs, downloads, and a way to talk to us land on this site as the product fills out.