Intelligent Agent Routing
EDDI's multi-agent system goes beyond simple load balancing. It provides intent-based agent discovery that automatically routes user requests to the most appropriate agent based on declared capabilities.
Orchestration Features
- Agent Discovery: Agents declare intents and are auto-matched to user requests
- Managed Conversations: One conversation per intent+user, auto-created
- Agent Triggers: Automated agent activation based on events
- A/B Routing: Test different agents against the same traffic
- Coordinator Pattern: Agents delegate to specialized sub-agents
Group Conversations & Advanced Orchestration
EDDI supports multi-agent group conversations with 7 built-in discussion styles, explicit voting, shared work products, standing teams, nested group structures, dynamic agent creation, and a meta-agent that operates the platform itself:
- 7 Discussion Styles: Round Table, Peer Review, Devil's Advocate, Delphi, Debate, Task Force (PLAN → EXECUTE → VERIFY → SYNTHESIS), and Negotiation (offers, concessions, and an arbiter)
- Voting: Vote phases collect explicit ballots (majority or approval, weighted, quorum-gated) and record a decision with the full tally and the losing side's dissent
- Shared Artifacts & Task Bidding: Members co-edit a validated shared work product and bid for tasks instead of being assigned round-robin
- Humans as Members: A person takes a real turn in the discussion, not just an approval
- Live Discussion Overview: A dashboard of a running discussion in the Manager and the Workforce board: each member's current stance, live cost, bids and directional turns, with a switcher for earlier rounds
- Decisions You Can Read: A vote that turns a proposal down ends as
REJECTED, not as a failure, and the editor says when debate roles turn a synthesis into a verdict - Standing Teams: Persistent teams with a backlog, cron cadences, a facilitator with bounded moves, and retro phases that harvest team memory
- Preset Templates: Research pod, editorial team, ops task force, decision board, and negotiation table, ready to instantiate
- Dynamic Agents: Agents in Task Force discussions can create, recruit, delegate to, and teardown sub-agents at runtime with guardrails (provider/model whitelists, per-discussion caps, lifecycle policies)
- Nested Groups: Compose groups of groups for tournament brackets, red-team vs blue-team, and panel reviews
- Slack Integration: Deploy agents to Slack channels and run multi-agent debates directly in threads, with DM support, trigger-keyword routing, and an observe mode where an agent watches a channel and replies only when a keyword or file type matches, within cooldown, daily reply, and daily cost limits
- Platform Operator: A meta-agent that reads and operates your deployment (including creating other agents), with every write behind a human approval gate
- A2A Protocol: Full Agent-to-Agent protocol implementation with Agent Cards and cross-platform skill discovery
- Capability Matching: Discover and route to agents by skill, confidence score, and custom attributes
Why vote rather than just debate? Research on multi-agent debate finds that majority voting accounts for most of its measured gains, and that one confident, persuasive agent can pull a group toward a wrong answer. EDDI's votes are cast independently of each other, record the losing side's dissent, and let each member run on a different model. The evidence page has the studies.