Controllable Agent
A multi-agent framework in which a planner, paired coder-and-reviewer teams, a coordinator and a memory-extracting agent divide up a complex task, work in parallel, and build a searchable knowledge base from what they learn.
Controllable Agent is a framework in which five agent roles cooperate on a task instead of one model handling it start to finish: a coordinator that manages the others without doing the work itself, a planner that writes and continually adjusts the plan, coder-and-reviewer pairs that write and immediately check each module, and a memory agent that records what happened and distills reusable skills from it.
Coder-and-reviewer pairs can run several at once, controlled by a concurrency limit, with unlimited retries at the module level and up to three rounds of integration testing before an issue is escalated to the user, who can interrupt at any point. After each task, a lightweight model call turns the conversation into a structured summary, which is indexed for keyword search and, once enough summaries accumulate on one topic, merged automatically into a longer wiki-style knowledge page, with a lightweight relationship index built alongside it.