SethGammon/Citadel — The operating layer for Claude Code + OpenAI Codex: persistent project memory, intent routing, safet…

Description and use

**Brief Summary (2‑3 sentences)**
SethGammon/Citadel is an open‑source orchestration layer that adds persistent memory, intent routing, safety hooks, cost telemetry, and parallel‑agent fleet management to Claude Code and OpenAI Codex. It turns ad‑hoc prompts and tool calls into repeatable, observable multi‑agent workflows, making it easy to build robust AI‑powered automation pipelines. With active development, 600+ stars and recent commits, it is ready for serious pilot projects.

**Value**

• **Workflow repeatability** – By persisting project state and routing intents, Citadel lets you treat a series of prompts and tool invocations as a single, version‑controlled pipeline.
• **Scalability & safety** – Parallel agent fleets can be spun up on demand, while built‑in safety hooks and cost telemetry keep execution under control.
• **Standardisation** – Provides a common API/SDK/CLI for memory handling and tool integration, reducing the engineering effort required to stitch together Claude, Codex, or other LLMs.

**Practical Adoption Path**

1. **Evaluate the API/SDK** – Clone the repo, run the provided CLI demo, and connect your existing Claude or Codex keys.
2. **Prototype a simple workflow** – Define a small “intent” (e.g., generate a function, run a linter, commit to Git) and let Citadel manage memory and routing.
3. **Scale to parallel agents** – Use the fleet manager to launch multiple agents for tasks such as batch code review or test generation.
4. **Integrate safety & telemetry** – Enable the built‑in hooks to enforce policy checks and capture per‑request cost data.
5. **Deploy** – Containerise the service (Dockerfile is included) and run it behind your internal API gateway for production use.

**Production Readiness**

• **Activity & Community** – 615 ★, 57 forks, recent commits (as of 2026‑06‑24) and a growing issue/PR discussion indicate an active maintainer base.
• **Maturity** – Core features (memory, routing, safety, telemetry) are stable; the JavaScript codebase is well‑documented and includes an SDK and CLI for easy integration.
• **Risk Profile** – No major metadata or licensing concerns identified, though a final security audit and confirmation of maintainer responsiveness are recommended before full‑scale rollout.

Overall, Citadel offers a production‑grade foundation for building repeatable, safe, and cost‑aware multi‑agent AI pipelines, making it a strong candidate for pilot deployments and eventual enterprise adoption.

Problem

SethGammon/Citadel helps turn isolated prompts and tools into repeatable agent workflows.

Outcome

High for an OSS candidate: recent activity, adoption, and ecosystem signals are strong enough for a serious pilot.

Workflow

• coordinate multi-agent workflows
• add tool-use pipelines
• standardize agent memory

Evidence

The operating layer for Claude Code + OpenAI Codex: persistent project memory, intent routing, safety hooks, cost telemetry, and parallel agent fleets.

Source