antonbabenko/deliberation — Ask Codex, Gemini, Grok, and 400+ OpenRouter models (Qwen, Kimi, DeepSeek) for second opinions or ar…

Description and use

**Brief Summary (2‑3 sentences)**
*deliberation* is an open‑source orchestration framework that lets you query dozens of LLMs—including Codex, Gemini, Grok, and 400+ OpenRouter models—and reach a consensus or “second‑opinion” verdict via an MCP (Model‑Consensus‑Protocol) server. It turns ad‑hoc prompts and tool calls into repeatable, multi‑agent workflows, measuring which models earn a seat at the table and exposing a simple API/CLI for integration.

**Value Proposition**

• **Unified multi‑model access** – One endpoint can invoke a wide palette of LLMs, making it easy to compare answers, aggregate expertise, or let a designated arbiter (Claude Code, Cursor, etc.) resolve conflicts.
• **Workflow repeatability** – By codifying prompt sequences, tool‑use steps, and memory handling, teams can ship deterministic agent pipelines rather than fragile one‑off scripts.
• **Performance insight** – Built‑in metrics track each model’s success rate, latency, and cost, helping you prune low‑performing models and keep the most valuable ones in production.

**Practical Adoption Path**

1. **Prototype** – Clone the repo, run the provided Docker compose or npm run dev to spin up the MCP server locally. Use the CLI (delib query …) to experiment with a few models on a sandbox dataset.
2. **Integrate** – Replace existing single‑model API calls with the deliberation SDK (JavaScript/Node) or REST endpoints. Define your consensus policy (e.g., majority vote, arbiter‑mediated) in a JSON config file.
3. **Scale** – Deploy the MCP server to a managed container service (K8s, Cloud Run). Hook up your CI/CD pipelines to automatically refresh the model list from OpenRouter and to monitor the built‑in metrics dashboard.
4. **Govern** – Use the provided model‑seat scoring to enforce cost caps or compliance rules, and persist the consensus outcomes in your existing data store for auditability.

**Production Readiness**

• **Activity & Community** – 105 ★, recent commits (last update 2026‑06‑23), and a growing set of topics indicate an active maintainer base.
• **Architecture** – The MCP server is a stateless Node.js service with clear API/CLI contracts, making it easy to containerize, load‑balance, and monitor.
• **Observability** – Built‑in logging and model‑performance metrics give operators the visibility needed for SLA enforcement.
• **Risk Assessment** – No glaring metadata or licensing issues have been identified, though a final security audit (dependency scanning, secret handling) and confirmation of maintainer responsiveness are recommended before mission‑critical rollout.

Overall, *deliberation* is a mature OSS candidate for teams that need to coordinate multiple LLMs, enforce consistent tool‑use patterns, and gain data‑driven insight into model effectiveness. It can be piloted quickly and scaled to production with modest engineering effort.

Problem

antonbabenko/deliberation 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

Ask Codex, Gemini, Grok, and 400+ OpenRouter models (Qwen, Kimi, DeepSeek) for second opinions or arbiter-mediated consensus. One MCP server for Claude Code, Codex, Cursor, Kiro, OpenCode. Measures which models earn their seat.

Source