JSON-Driven Agent Logic
In EDDI, everything that defines an agent's behavior is a JSON configuration resource: rules (system prompts), LLM connector settings, workflow pipelines, tool configurations, and deployment descriptors. No code changes, no redeployment.
Why This Matters
- No redeployment: Change agent behavior by updating JSON, not source code
- Prompt engineer friendly: Non-developers modify rules via UI or REST API
- Version controlled: JSON configs live in Git alongside infrastructure-as-code
- Hot-reloadable: Changes take effect immediately, no server restart needed
- Auditable: Every configuration change is tracked in the immutable audit trail
- Import / Export: Agents are portable as ZIP files with automatic secret scrubbing on export. Import merges or overwrites, no version conflicts
- Agent Sync: Live instance-to-instance sync with structural matching, content diffing, and selective resource picking, no ZIP intermediary needed
- Prompt Snippets: Reusable, versioned system prompt building blocks referenced as
{{snippets.safety_rules}}, compose prompts from shared libraries - Behavior Rules: IF-THEN logic engine for routing, orchestration, and business logic decisions without code
- Version Following: Mark a new agent version as compatible with the previous one, and open conversations continue on it instead of being ended. The Manager shows what each deploy does to the conversations already running
Resource Types
EDDI manages these configuration resources: Rules (system prompts), Workflows (pipeline definitions), LLM Connectors (provider settings), HTTP Connectors (external API integrations), Connections (how to authenticate to an external system), Groups (multi-agent discussions), Agents (deployment descriptors), and Environments (deployment targets).