Explore research and engineering articles from the MiniMax Code team on agents, coding workflows, and Mini Apps.
2026-11-05
Why You Should Be in AI, and How to Do It
Agents made experiments almost free. Now the scarce skill is asking the right question and designing the right control. Curiosity, execution and scientific literacy, and why they stack.
Perspective
From Personal Agent to Agent Team: Designing and Open-Sourcing the BYOB Collaboration Protocol
On a trusting team, people end up relaying between bots. We put everyone’s bots at one table: a shared ledger, seats, the floor and a commit gateway. The protocol and a reference implementation are open source.
Agent Collaboration
2026-10-29
MCode Engineering Notes 05 | How Does the Agent Know the Current Time?
The model has no clock. The most direct fix is to inject the time into the system prompt — but a time that changes every turn can trash the prefix cache. From the intuition through why the prefix matters to how MCode does it.
MCode Engineering Notes
2026-10-27
MCode Engineering Notes 04 | How Long Should an Agent Wait on a Bash Command? From Foreground Blocking to Background Tasks
When an agent runs a bash command, should it wait for it to finish? Following that question from foreground blocking all the way to background tasks, plus the notifications, anti-polling, cleanup, and post-compaction reminders that come with them.
MCode Engineering Notes
MCode Engineering Notes 03 | Why Long Sessions Make Agents Dumber: Compaction vs. Memory
What to do as sessions get longer: what compaction and memory are each responsible for, where each one breaks, and how they fit together.
MCode Engineering Notes
2026-10-26
MCode Engineering Notes 02 | Too Many Tools Will Eat Your Context: From Tool Search to PTC
In the order we actually took in MiniMax Code: cut first, then add tool search, and finally move to PTC (letting the model write programs that call tools).
MCode Engineering Notes
2026-10-25
MCode Engineering Notes 01 | More Subagents Isn’t Better: Manage Context First, Then Parallelize
Subagents are for managing context, not for carving up territory: when to split, when not to, and how to fold them back in once you’ve split too much.
MCode Engineering Notes
2026-10-20
Product Diary: Building Professional-Grade Mini Apps on Agent OS
From chatbots and workflows to agents, and now Mini Apps: why we chose to build the agent into the computer as a system service.
Mini App
Introducing MiniMax Code Mini App
A Mini App is a long-running application with a frontend and a backend. It installs as a plugin, calls and is called by the Agent, and keeps the Agent’s work on a page that stays live.
Mini App
2026-10-08
I stopped babysitting my coding agents. Here’s the workflow.
What an audit of my multi-agent setup changed, and how I now define, verify and deliver work with core-grill, core-spec and deliver.
Agent Workflow
2026-09-22
Building a Testbed for Coding Agents
MiniMax Code Engineering Notes, Part 2: task bundles, the Runner, independent verification, compaction experiments, and A/B comparison — how to know a Harness change really made things better.
Agent Evaluation
2026-09-19
MiniMax Code Is Open Source: Inside the Harness Behind a Coding Agent
MiniMax Code Engineering Notes, Part 1: capability assembly, context compaction, tool execution eligibility, canonical history commits, and sub-task delivery.
Agent Harness