Relentless: the orchestrator that never stops

We published Relentless on npm on January 12, 2026: @arvorco/relentless, version 0.1.0. It’s a universal orchestrator for coding agents — it takes whatever agent you already use (Claude Code, Codex, Gemini, whatever comes next) and puts it to work in a loop until the task list is empty.

The idea came from a problem anyone who has used a coding agent for more than an afternoon recognizes: you hand it a big task, it works well for a while, and then it starts repeating steps, losing the thread, or stalling on decisions it should have already made. Long context degrades. It always has.

Relentless fixes this without magic: every iteration starts with fresh context. No dragging one endless conversation until the agent forgets what it was doing. Memory between iterations doesn’t live in an increasingly polluted context window — it lives in git. Commit by commit, the repository’s history is the agent’s memory. Simple, auditable, and it works with whatever version control you already use.

How it works

You define a task list — an implementation plan, a bug backlog, an entire migration. Relentless invokes the agent you picked, lets it work one task, validates the result, commits, and starts the next iteration with a clean slate. No babysitting. No pasting the same prompt back in at 2am because the agent wandered off.

This works because most of the cost of a coding agent isn’t the task itself — it’s the accumulated context it drags along while making progressively worse decisions. Resetting context at every step, backed by reliable external memory, is both cheaper and more predictable.

Smart Auto Mode: real savings, not a promise

Relentless ships with what we call Smart Auto Mode. It automatically decides when a more expensive model is worth it versus a cheaper one, based on the actual complexity of the task at hand. In practice, that’s produced 50% to 75% savings on squads we’ve already run in production — including an internal squad of ours pairing Relentless, Claude Code, and Codex on the same pipeline, one of the case studies we use in our agentic consulting work.

This isn’t a cosmetic optimization. It’s the difference between running an overnight code migration for the price of dinner versus the price of a month of SaaS seats.

Traction in month one

We shipped it, documented it, and let it run. A month later: roughly 80 GitHub stars and 4.4k npm downloads per month. Not unicorn numbers — numbers of a tool that solves a real problem for people who already live inside a terminal running a coding agent all day.

Why this matters for Arvor

I spent two decades building enterprise search engines — Solr, Lucene, systems that had to be reliable in production, not just impressive in a demo. Coding agents have the exact same problem search always had: they look great in a demo and fall apart at scale if you don’t build the right infrastructure underneath.

Relentless is open source because Arvor isn’t in the business of hiding the pipeline that makes agents actually work in production. We’re in the business of building that pipeline, documenting it, and helping companies run the serious version of it. If you’ve already hit the wall of degrading context, or want to know how to plug this into your own operation, talk to us.