Agent Loop Python Packages

protocore

Protocore — the agent loop, and nothing else. A protocol-first ReAct runtime for LLM agents in Python 3.12+: context budget, tool surface, compaction, snapshot and resume. No database driver, no HTTP endpoint — everything outward is a Protocol you implement. MPL-2.0.

loop-memory

Local-first, agent-agnostic memory system that closes the agent loop on any AI agent. Watches any agent's transcript directory on disk and distils long sessions into a curated wiki that re-injects into the next turn. Out of the box: hooks for Codex / Claude / Hermes / OpenClaw; SDK + generic watcher CLI for the rest.

millrace-ai

Turns repeatable, domain-agnostic workflows into graph-driven loops.

brrragent

Reusable agent loop with multi-provider routing and MCP tool calling

bumblehive

An agent runtime designed for tool-driven loop engineering.

agents-harness

Declarative scheduled flow runner, per-request agent loop, and job kernel for local agent services (MCP + CLI).

agentloop-cli

AgentLoop — harness-agnostic, self-verifying autonomy wrapper for coding agents. Drives OpenCode/Kilo/Claude/Aider/Codex in a verify-gated loop with a held-out oracle, so the agent actually finishes and proves it is correct.

loomc

loomc is a Python decorator that turns a single-iteration function into a resilient agent loop — with hard budget caps, SQLite checkpointing after every step, and full resume from wherever it stopped.

foreman-orchestrator

A Boris-style agentic orchestrator TUI that supervises headless Claude Code agents through a gated software-delivery pipeline — pointed at any repository.

judge-drift-sentinel

Attribute eval-score movement to the system or the LLM judge using a frozen human anchor set and kappa.

ralph-harness

A repo-native coding-agent loop harness for Claude, Codex, Copilot, and other CLI agents. Agents can edit. Gates decide what lands. You set the plan in motion. The agents take a prompt, and each iteration must update specs and commit through guardrails.