EDDI

EDDI is an open source platform for deploying governed AI agent systems inside an organisation's own infrastructure. It is built as a deployable orchestrator rather than a library: agent logic, pipelines and tool definitions live as JSON configuration, so prompt engineers can change behaviour through a React management interface or a REST API without redeploying code. The current major version is v6, with the 6.4.0 release adding per user workspaces, governed outbound Connections and a Red Hat certified container image. The platform targets enterprises in regulated settings that need self hosting, auditability and human oversight. It exposes 84 tools over the Model Context Protocol, so a client such as Claude Desktop can operate it, and its agents can consume external MCP tools too. Multiple agents can be grouped into deliberation formats such as round tables, structured debates, Delphi rounds and votes, with a human signing off. Approval gates can pause a whole conversation or a single tool action, requests can be approved from Slack, and pending approvals survive a restart. Every pipeline step is written to an immutable audit ledger with tokens, cost, timing and tool calls. Other features include vault based secrets, retrieval over seven embedding providers and six vector stores, model cascading for cost control, scheduling, and an API compatible with the OpenAI protocol for use behind Open WebUI or LangChain. The core platform is free to self host from GitHub. A managed cloud offering and enterprise support are sold through the separate EDDI Solutions site, and no prices are published for them, so pricing is quote based. The project positions itself explicitly against prototyping tools such as Flowise and n8n and against forcing agents into deterministic workflow engines, arguing that visual prototypes should carry straight through to production.