50-100+ agents across multiple environments. Coordination, governance, and scaling — built into the engine.
When you need more than one agent, you need orchestration. Not a layer on top of 5 different tools — an engine that runs agents AND coordinates them. Stigmergy-based coordination, agent deliberation councils, automatic review and merge, drift tracking, and guardrails.
Traditional multi-agent systems use direct message passing — fragile and doesn't scale. Codebolt uses stigmergy: agents communicate through structured shared state. Pheromone-based task ownership prevents conflicts. The engine manages sequencing, dependencies, and resource allocation.
Agent A → shared state ← Agent B
↕
Agent C → shared state ← Agent D
Pheromones: ownership signals
No direct messaging needed
Scales: 2 agents → 200 agents
The provider system lets one Codebolt instance treat another as a coordinated environment. Run agents across local, Docker, cloud, and hierarchical setups. Each environment has full lifecycle management with heartbeat monitoring. Git bundles transfer state between environments.
Main Instance
│
┌─────┼──────┐
↓ ↓ ↓
local docker cloud
│ │ │
└──────┼──────┘
git bundle state transfer
heartbeat monitoring
Paperclip orchestrates agents but doesn't run them. You need to bring Claude Code, OpenClaw, Cursor, and wire them together. Codebolt IS the engine — it runs agents AND orchestrates them. One system instead of five tools stitched together. And every agent has access to the full 63-module SDK, not just whatever each external tool exposes.
Paperclip:
✓ Org charts, budgets, governance
✗ Doesn't run agents
✗ Requires external tools
✗ Limited by each tool's API
Codebolt Orchestration:
✓ Runs agents (63-module SDK)
✓ Coordinates (stigmergy)
✓ Governance (guardrails)
✓ One engine, not five tools
Agents propose options, debate tradeoffs, vote, and reach consensus. Review-merge requests let agents review each other's work without GitHub PR bottleneck. Conflict detection runs before merge. Drift tracking catches deviation across all agents. Every decision is logged and traceable.
Agent A: implements feature
Agent B: reviews code
Agent C: checks for conflicts
Agent D: runs tests
↓
Auto-merged. Decision logged.
No human bottleneck needed.