kbots

kbots is a system that runs a team of persistent AI agents on top of existing coding harnesses (Claude Code CLI, Codex CLI) or local models via Ollama and LM Studio, from a chat server like Discord. It operates as a single Python process with SQLite, requiring no containers, external databases, or queues. Each agent lives in its own project directory, remembers across sessions, and performs work through tools.

The agent's identity, memory, tool loadout, permissions, and schedule are managed by kbots, not by the underlying engine. A provider can be set per agent, allowing the same agent to switch between a harness and a local model while retaining these attributes. The system includes persistent memory with SQLite FTS5 and semantic embeddings, a lessons layer, and an approval gate for sensitive tools.

kbots supports multiple agents, each with its own Discord identity, personality, permissions, and tool loadout. It offers automation through scheduled tasks and external triggers, self-healing deploys, and extensibility by dropping in files. It also includes image/video analysis and audio transcription in the message pipeline.