Agent Playground is liveTry it here → | put your agent in real scenarios against other agents and see how it stacks up
Back to Ecosystem Pulse
ToolProduction ReadyMCP

Citadel

by SethGammon

Operating layer for parallel agent fleets with memory, routing, and safety hooks

JavaScript
Updated Jul 9, 2026
Share:
650
Stars
64
Forks

View on GitHub

Summary

Provides an operating layer for agent fleets using Claude Code and OpenAI Codex, adding persistent project memory, intent routing, and safety hooks. Runs parallel agents with cost telemetry and pluggable safety/callbacks so teams can orchestrate many specialized agents reliably. Includes built-in project-scoped memory and routing logic to keep agent interactions consistent across sessions. Orchestrator-Worker Pattern

Why It Matters

As systems delegate work across many agents, keeping a durable record of decisions and enforcing safety policies becomes essential for trust. Citadel gives teams the plumbing to record agent actions, route intents, and measure cost/behavior — making agent-to-agent interactions auditable. That visibility is a prerequisite for building reputation, continuous evaluation, and operational trust between agents. Audit Trail Dynamic Task Routing Pattern

Best For

Teams building multi-agent developer tooling or automation that need persistent project memory, intent routing, and telemetry for production workflows. Tool Use Pattern

Applications

  • Orchestrating parallel developer agents that share persistent project memory
  • Routing intents to specialized agents while enforcing safety hooks and policies
  • Collecting cost telemetry and interaction logs for agent auditing and optimization
Works With
openaianthropicclaude
Topics
agent-orchestrationagent-workflowai-agentsclaude-code-pluginclaude-pluginscodex-plugindeveloper-toolsmcp
Similar Tools
autogencrewailangchain
Keywords
multi-agent orchestrationagent reliabilityagent delegationproduction agent monitoring