Now in platform preview

The swarm operates. You stay in control.

repo-boss is an AI-powered repository governance and engineering operations platform built around policy-bounded agent execution. It inventories GitHub organizations, monitors repo health, supports PR review and code alerts, tracks deployments, and explains how engineering work is routed.

Repos · PRs · Deployments · Routing · Policy · Audit

repo-boss · Operations
live

4

Runs in flight

94%

Success / 1h

81%

Max saturation

runningTriage elevated 429s on github connector
runningRebalance connector-ops capacity
succeededDraft Q3 saturation report
blockedPlatform reset (destructive)

destructive actions blocked by policy · every dispatch scored & logged

How it works

Govern the repo work. Route the agent work.

repo-boss is not a generic agent demo or a task queue with a dashboard bolted on. It is a governed control plane for repository operations, engineering work, connector execution, and policy-bounded automation.

01

Connect repository and engineering signals

repo-boss brings GitHub org inventory, repository health, code alerts, PR activity, deployment events, incidents, and developer metrics into one governed engineering operations surface.

02

Route work through the control plane

Tasks are routed through a deterministic scoring engine that evaluates capability coverage, tool support, authority scope, agent health, budget headroom, and runtime capacity before work is queued or blocked.

03

Inspect the decision and the outcome

Every routed or blocked run carries a trace: selected swarm, selected agent, candidate scores, policy decision, reasoning, execution status, cost, tokens, and audit history.

Features

Repository operations need a cockpit.

Built for engineering operators who need repo intelligence, agent routing, connector governance, and owner controls in the same operating layer.

Core

Repository governance

Inventory GitHub organizations, surface repository health, ingest code alerts, support PR review posting, track deployment activity, and organize governance work into Forge projects.

Deterministic route traces

Every run records candidate scores and reasons across capability, tool class, authority, health, budget, and runtime capacity. Operators can explain why work moved or why it was blocked.

Ralph engineering orchestrator

Ralph gives tenant-scoped engineering teams a two-pass conversational interface for DORA metrics, repo health, code alerts, active incidents, developer metrics, and Atlas context when needed.

Platform Owner Console

Owner routes provide cross-tenant fleet overview, tenant drill-down, AI quota management, active workspace spend, trust posture, and an owner copilot for proposed platform actions.

MCP Boss execution path

Connector execution goes through MCP Boss for catalog discovery, fnmatch policies, approval queues, per-tool rate limits, execution logs, and shared audit events.

Operational analytics

Track autonomy rate, recovery rate, task completion, token and cost usage, incident posture, runtime saturation, connector latency, and routing candidates evaluated.

Routing engine

Six dimensions. One decision. Full reasoning.

When repo-boss routes work, it scores every candidate agent across capability match, requested tool class, authority scope, live health, budget headroom, and runtime capacity.

The result is not a mystery handoff. The platform records which agent won, which agents were blocked, what each candidate scored, and whether policy allowed execution.

Routing decision · candidate scoringallowed
winnerIncident Responder+91
Capa
Tool
Auth
Heal
Budg
Runt
Executive Orchestrator+73
Capa
Tool
Auth
Heal
Budg
Runt

Swarms & agents

Six swarms. Thirty-five agents.

Work is organized into domain swarms, each with specialist agents, runtimes, capacity, budgets, and authority scopes — coordinated by the control plane.

Command Swarm

Cross-platform planning, routing, incident response, documentation, security review, and budget governance.

Connector Swarm

External API execution, connector recovery, credential checks, rate-limit handling, and governed write-external work.

Asset Swarm

Asset registration, memory placement, lineage, retrieval operations, and knowledge curation.

Analytics Swarm

Telemetry interpretation, evaluation scoring, anomaly detection, drift detection, and health explainability.

Engineering Swarm

Specialized platform, database, frontend, integration, DevOps, data, security, testing, auth, and LLM engineering work.

Repo Governance Swarm

GitHub organization inventory, repository discovery, code analysis, dependency audit, PR review, conflict handling, and deployment tracking.

For engineering leaders

The difference between “we have AI agents” and “we have governed engineering operations.”

A repo dashboard can show what happened. A governed engineering control plane can route the next action, score the decision, recover failures, replay runs, surface incidents, and prove what changed.

Route

Deterministic explainable routing — capability, tool support, authority, health, budget, and runtime capacity recorded on every run.

Recover

Replay failed runs, acknowledge incidents, recover stale work, and keep runtime health visible in the operator console.

Prove

Audit every dispatch, connector execution, approval, blocked action, and outcome. Governance is inspectable, not assumed.

Governed execution

Autonomy with the boundaries built in.

Authority scope

Agents are gated by action classes such as read, write-safe, and write-external before the route can execute.

MCP Boss policy

Connector calls pass through catalog lookup, fnmatch policy checks, approval requirements, rate limits, execution logs, and audit events.

Owner controls

Fleet, quota, spend, trust, and tenant lifecycle operations sit behind owner/platform-admin routes and proposed-action confirmation flows.

Honest failure

When an executor, provider, runtime, or approval path is unavailable, repo-boss returns an explicit blocked or failed reason instead of pretending the action worked.

Govern repository operations with a real control plane.

Connect repo signals, route agent work through policy, inspect every decision, and keep engineering autonomy observable before it becomes operational risk.

MCP approvals · Audit logs · MFA