Roadmap
kipp•punkt is under active development. Here is what’s planned next.
Task-specific short-term memory
Section titled “Task-specific short-term memory”Agents currently start each task from scratch. Short-term memory gives an agent context about what it already tried within a single task, so it avoids repeating mistakes and builds on prior attempts.
Collective long-term memory
Section titled “Collective long-term memory”Learnings from one task should carry over to future tasks. Long-term memory captures patterns, preferences, and recurring feedback across the full history of a repository, so agents get better over time without you repeating yourself.
Multi-repo orchestration
Section titled “Multi-repo orchestration”Running separate kipp•punkt instances per repository works but scales poorly against GitHub rate limits. A centralized orchestrator that manages multiple repositories from a single process will remove this bottleneck.
Periodic refactoring agent
Section titled “Periodic refactoring agent”After several features land, codebases accumulate drift. A refactoring agent that runs every n cycles reviews recent changes, identifies structural debt, and opens cleanup PRs automatically.
Improved dependency scheduling
Section titled “Improved dependency scheduling”Today, task ordering is mostly manual. Better dependency scheduling will let the orchestrator reason about which tasks block others and sequence work accordingly, especially in multi-agent setups.
Voice calls on GitHub issues
Section titled “Voice calls on GitHub issues”Complex ideas are easier to talk through than type out. The goal is to let you start a voice conversation with the kipp•punkt agent directly from a GitHub issue, using your existing call infrastructure, for faster refinement on topics where text is too slow.