RAPTOR — Open-Source Multimodal AI Framework & Agent Harness

Open-Source Multimodal AI Framework & Agent Harness An enterprise AI runtime for building agentic applications over video, audio, images, documents, and organizational knowledge

About • Features • Installation • Quick Start • Documentation • Contributing • Website

---

RAPTOR is an enterprise AI runtime for building agentic applications over video, audio, images, documents, and organizational knowledge — providing MCP tools, A2A orchestration, persistent memory, hybrid RAG, GraphRAG, evaluation, guardrails, and model lifecycle services. Developed by the DHT Taiwan Team at DHT Solutions.

🚀 Current Release

Aigle 0.4 - Community Beta (August 2026)

This release continues the open-source RAPTOR framework, codenamed “Aigle”. Release 0.4 grows the platform to 27 independently deployable Docker Compose modules (three earlier modules — hybrid-search, graph-database, graph-service — are retired and kept only for rollback) driven by the same single build system (Aigle/0.4/deployment/modules/build.py), and delivers the v0.4 roadmap: MCP interfaces across core services, persistent multimodal memory, and a per-user isolated multi-model database that replaces the previous Qdrant/OpenSearch/Neo4j trio.

Highlights:

• 🔌 MCP Server — Raptor’s search, RAG, asset, and memory APIs exposed as Model Context Protocol tools/resources for LLM clients (module 27)
• 🧠 Memory Service — persistent per-user/per-session memory (facts, preferences, conversation history, multimedia) backed by MemVID, layered underneath the existing Redis short-term chat cache (module 26)