Skip to main content

OPENDEV: Terminal-Native AI Coding Agents

1 min read

OPENDEV: Terminal-Native AI Coding Agents

IDE plugins are the wrong shape for AI coding. Real development work lives in the terminal, in source control, in builds and deployments, and a plugin-based agent loses its context the moment you leave the editor.

OPENDEV is an open-source CLI agent built for that shift.

The architecture

A dual-agent design separates the planner from the executor, which prevents runaway actions. Workload-specialised routing sends each task to an appropriate model instead of using one model for everything. Tool discovery is lazy, loading tools only when they are needed, which cuts cold-start time.

The context problem

Long coding sessions bloat context, and reasoning degrades along with it. OPENDEV applies adaptive context compaction, progressively compressing older observations while keeping recent work at full fidelity.

An automated memory system accumulates project-specific knowledge across sessions. It is paired with event-driven system reminders that counteract instruction fade-out, the slow drift where a model forgets its directives partway through a long task.

Limitations

It is work in progress and shows it. Terminal-only, so there is no visual feedback. The model routing adds real complexity for the operator to reason about.

What is worth taking

Context efficiency matters more than raw model capability for long-horizon work. The dual-agent split, lazy tool loading and compaction strategy are transferable whether or not you use OPENDEV itself.

Paper: https://arxiv.org/abs/2603.05344

Share
XLinkedIn

Stay in the loop

One dispatch per week — what I shipped, what broke, and what I learned from the field. No filler.

By subscribing, you agree to receive occasional emails. You can unsubscribe at any time.

What should I write about?

Got a topic you'd like me to cover? I read every suggestion.

More in Blog