Chairman LLM

AI agent · Workflow / back-office agent · Operations & supply chain by Blackbox AI

The orchestrator and judge for multi-agent coding workflows.

It automates the promotion of the cleanest branch to a pull request while maintaining a transparent audit trail.

Agent specs

The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.

Languages
English

Category: AI Agents

Channels
API; Dashboard
Integrations
GitHub; GitLab; Bitbucket; Blackbox Agents API
Models
Microsoft Decision-1
Deployment
Cloud; Single-tenant isolated deployment (Enterprise)
Role
Multi-agent orchestrator and judge
Evaluation criteria
Correctness, scope, risk, clarity
  • Ranks implementations from up to 4 parallel agents
  • 73.6% classification accuracy on held-out repositories
  • Automates pull request selection and promotion
  • Provides a transparent decision trail for audits

Overview

Chairman LLM is an orchestration and judging agent that manages the output of multiple coding agents working in parallel. It is designed for enterprise engineering teams to automate the selection of high-quality code implementations from competing model attempts. The agent evaluates branches for correctness, scope, risk, and clarity, automatically promoting the cleanest result to a pull request. It provides a transparent decision trail for every implementation choice, allowing for technical audits of the reasoning process. While it automates the selection, human users retain the ability to override its decisions through a central dashboard. Chairman LLM is integrated into the Blackbox Agents API and supports orchestration across models like Claude, GPT, and Gemini.

Orchestration and Judging

  • Implementation ranking: The agent reviews competing implementations from multiple models for correctness, scope, risk, and clarity.
  • Audit trail generation: The agent provides a decision trail explaining why a specific implementation was selected before it is merged.
  • Risk assessment: It evaluates branches for potential risks and adherence to the defined task scope.

Human Control and Visibility

  • Human override: Humans can override the selection made by Chairman LLM from the dashboard.
  • Dashboard monitoring: Teams can monitor all agents in one dashboard and compare implementations side by side.

What it does

  • Rank implementations
  • Select best result
  • Review branches for correctness
  • Evaluate risk and scope
  • Generate decision trails

Published results

  • 73.6% tier classification accuracy on held-out repositories
  • 65% task completion on Terminal-Bench 4.0 when used in multi-agent router

Use cases

  • Platform teams use the decision trail to audit why specific code changes were selected for production pull requests.
  • Developers automate the promotion of sandboxed agent work to reviewable Git branches based on correctness and risk scores.

Specifications

Channels
API; Dashboard
Integrations
GitHub; GitLab; Bitbucket; Blackbox Agents API
Human control
Humans can override the selection from Chairman LLM in the dashboard
Models
Microsoft Decision-1
Deployment
Cloud; Single-tenant isolated deployment (Enterprise)
Role
Multi-agent orchestrator and judge
Evaluation criteria
Correctness, scope, risk, clarity
Integration interface
Blackbox Agents API (HTTP)
VCS support
GitHub, GitLab, Bitbucket
Maximum parallel agents
4 per task
Routing model
Microsoft Decision-1
Decision time
~0.5 s end-to-end through gateway

How it works

  1. Launch parallel agents: A single task is sent to multiple agents, such as Blackbox, Claude Code, Codex, and Gemini, to work in parallel.
  2. Evaluate implementations: Chairman LLM reviews the competing branches for correctness, scope, risk, and clarity.
  3. Promote winning branch: The agent selects the winning branch and promotes it to a scoped pull request with a run summary and test evidence.

FAQ

How is the agent licensed?

The agent is included with an Enterprise agreement, which uses per-token commits that decrease as teams consume them.

Can I override the agent's decision?

Yes, humans can override the selection from Chairman LLM through the Blackbox dashboard.

What criteria does the agent use to judge code?

It reviews competing branches for correctness, scope, risk, and clarity before promoting the cleanest branch to review.

Added by AntSpace from public sources (blackbox.ai).

Related

More AI agents by Blackbox AI