Overview
CONF is a multi-agent conferencing framework where Claude Code agents collaborate autonomously or under user direction, with AfterLife Reincarnation that transparently respawns agents before they hit context limits — preserving full working memory across infinite sessions.
The problem: AI coding agents hit a hard wall — the context window. Once an agent fills its context with code, conversation, and tool results, it loses track of earlier work. Long-running tasks stall, and users have to manually re-brief new sessions. Multi-agent coordination is even worse: agents can’t talk to each other, share state, or pick up where a teammate left off.
The solution: CONF provides two modes of agent collaboration. In user-led mode, a human orchestrates multiple Claude Code agents, each specializing in different parts of a codebase, coordinating via structured messaging. In fully autonomous mode, agents run conference sessions independently — discussing architecture, reviewing each other’s code, and dividing work — with the user observing or stepping in as needed.
AfterLife Reincarnation: The Breakthrough
When an agent’s context usage approaches a user-configurable threshold (e.g., 70% of tokens), the system automatically:
- Saves the agent’s working state to persistent memory (CC_Mem)
- Spawns a fresh agent with full context restoration
- Seamlessly continues the task
The dying agent’s last act is writing a comprehensive handoff — what was done, what’s next, what decisions were made. The new agent picks up exactly where the old one left off, with zero user intervention.
Teams of AI agents can work for hours on complex tasks without context degradation. Each agent stays sharp because it’s never more than 70% through its context window.
Key Features
- User-led Conferencing - Orchestrate multiple agents with structured message passing between them
- Fully Autonomous Mode - Agents run conferences independently, dividing and conquering complex tasks
- AfterLife Reincarnation - Automatic agent respawn at configurable context threshold (default 70%)
- CC_Mem Persistence - Working state survives across reincarnations via keyword-indexed memory
- gmsg Messaging - Structured inter-agent communication with sender/receiver routing
- Conference Tagging - Agents join/leave named conferences, messages auto-route to participants
- Configurable Autonomy - Users set how much independence agents get, from fully supervised to fully autonomous
Use Cases
Large-Scale Refactoring
Tasks that exceed a single context window. Agents coordinate across multiple modules, frontend/backend/tests, with each agent maintaining full context through reincarnation.
Multi-Module Features
Coordinated changes across frontend, backend, and tests. One agent per module, all communicating through structured messaging, all immortal through AfterLife.
Code Review Conferences
Multiple agents review each other’s work. Architectural discussions between specialized agents (frontend expert, backend expert, security reviewer).
Long-Running Development Sessions
Hours or days of continuous work without context loss. Agents reincarnate as needed, maintaining full project history.
AI Pair-Programming Across Entire Codebase
Multiple agents working simultaneously on different parts of your codebase, coordinating changes, avoiding conflicts.
Architecture
┌──────────────────────────────────────┐
│ CONF Framework │
│ │
│ ┌────────────┐ ┌────────────┐ │
│ │ Agent A │───▶│ gmsg │ │
│ │ (70% ctx) │ │ Message │ │
│ └────────────┘ │ Broker │ │
│ └──────┬─────┘ │
│ ┌────────────┐ │ │
│ │ Agent B │◀──────────┘ │
│ │ (45% ctx) │ │
│ └────────────┘ │
│ │
│ ┌────────────────────────────────┐ │
│ │ AfterLife Monitor │ │
│ │ • Tracks context usage │ │
│ │ • Triggers reincarnation │ │
│ │ • Manages handoffs │ │
│ └────────────────────────────────┘ │
└──────────────────────────────────────┘
│
┌────▼─────┐
│ CC_Mem │
│ (SQLite) │
└──────────┘
What Makes It Different
vs. Manual Agent Orchestration:
- Automatic context management - No manual session restarts or re-briefing
- Persistent memory - Full state preservation across reincarnations
- Structured messaging - Agents coordinate through formal protocols, not ad-hoc prompts
vs. Single-Agent Workflows:
- Specialization - Each agent focuses on its domain (frontend, backend, tests)
- Parallelization - Multiple agents work simultaneously
- Immortality - No task too large, agents never “die”
vs. Other Multi-Agent Frameworks:
- Context immortality - AfterLife Reincarnation is unique to CONF
- Production-proven - Not research, actively used for real development
- Claude Code native - Built specifically for Claude Code workflows
Tech Stack
- Python - Core framework
- Claude Code CLI - Agent runtime
- MCP (Model Context Protocol) - Tool integration
- gmsg - Messaging infrastructure
- CC_Mem - Persistent memory (SQLite + FTS5)
- PyQt6 - IATV integration for visual management
Status
Beta — Actively used internally at Silver Wizard Software for daily development across 8+ projects simultaneously.
Internal Infrastructure: CONF is production infrastructure developed for Silver Wizard Software’s multi-agent development workflows. Available for licensing to organizations interested in autonomous AI agent systems.
Best Practices
- Set context threshold conservatively - 70% is recommended, allows buffer for handoff
- Use checkpoint saves - Mark critical state points explicitly
- Monitor agent conferences - Watch coordination patterns, intervene when agents diverge
- Start supervised, graduate to autonomous - Learn your agents’ behavior before full autonomy
Licensing
CONF is internal production infrastructure available for licensing to organizations building agentic AI systems.
Contact: silver.wizard001@proton.me for licensing inquiries.
Note: Repository is currently private. Available for licensing to qualified organizations.
Documentation
Support
- Email: support@silverwizard.dev
- Licensing: silver.wizard001@proton.me