Practical overview
When it fits
Use RAPTOR when you are building an application that needs a self-hosted runtime for multimodal content processing and agent-oriented retrieval, rather than looking for a ready-made end-user assistant.
Who it is for
- Developers building agentic applications that work across media files and organizational documents.
- Teams that need to compose search, RAG, memory and agent-interoperability services into a self-hosted runtime.
Documented uses
- Process uploaded media into transcripts, OCR, frame descriptions, summaries, embeddings and knowledge-graph data.
- Search content using BM25, vector and graph-based retrieval, with reranking in the documented video-search workflow.
- Build question-answering applications over uploaded organizational content using RAG chat.
- Expose search, RAG, asset and memory capabilities as MCP tools and coordinate agents through A2A.
- Benchmark application pipelines and add configurable input/output guardrails.
Requirements & limits
The repository is public and identifies RAPTOR as open source under the Apache-2.0 license. The Aigle 0.4 documentation labels the release a community beta. A development evaluation API is also described for testing without setting up infrastructure.
- The documented Aigle 0.4 deployment is infrastructure-heavy, with Docker, Compose, Python, an inference server and NFSv4 storage prerequisites.
- GPU-based media processing has specific NVIDIA requirements; the installation notes specify at least 24 GB VRAM and recommend 36 GB or more for video processing.
- The repository documents a 27-module Docker Compose architecture, so deployment involves configuring and operating services rather than installing a single standalone application.
- The documented upload-and-query flow requires SSO authentication and a bearer token.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Deploy RAPTOR Aigle 0.4 and prepare its API
Configure and start the documented RAPTOR modules, then access the API gateway documentation.
Before you start
- Docker Engine 24.0+ and Docker Compose v2.20+
- Python 3.10+
- An Ollama server reachable from the cluster; vLLM is an optional alternative backend
- NFSv4 storage mounted by the relevant services
- For GPU media services: NVIDIA Container Toolkit, a driver supporting CUDA 12.8+, and at least 24 GB VRAM
Steps
Clone the RAPTOR repository and change to RAPTOR/Aigle/0.4.
In deployment/modules, copy .env.example to .env and copy available module-specific .env.example files to .env; fill in hosts, credentials and model names.
From Aigle/0.4, build the shared GPU base image with bash deploy.sh -m 08 --build.
Start modules in dependency order with bash deploy.sh; use --cpu-only to skip GPU modules.
Open the gateway documentation at http://:8012/docs.
Expected outcome
A configured RAPTOR deployment with the API gateway documentation reachable at the host address.
Guide limitations
- The deployment also requires networked hosts, synchronized clocks and firewall openings specified in the build guide.
- This setup guide does not cover user login or application requests; the quick-start documentation describes those later steps.
Upload and query media through the RAPTOR API
Follow the documented API sequence to submit a media file and ask a RAG question about its contents.
Before you start
- A deployed RAPTOR Aigle 0.4 platform and API gateway
- A user account managed through the authentication module
- A media file to upload
Steps
Request an SSO token from the gateway's /api/0.4/sso/login endpoint.
Upload the file to /api/0.4/asset/fileupload_analysis using the token.
Search indexed content through /api/0.4/search/video_search, or send a question to /api/0.4/chat/chat with the token.
Expected outcome
The documented workflow submits media for AI processing and then supports search or a RAG chat question about the uploaded content.
Guide limitations
- This workflow depends on a deployed and configured platform; the repository separately identifies a development evaluation API.
- The documented endpoints require an authenticated bearer token.
Compare options for Agent orchestration
These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.
| Project | Documented purpose | Platform | Access and requirements |
|---|---|---|---|
| Raptor — DHT-AI-Studio | Process uploaded media into transcripts, OCR, frame descriptions, summaries, embeddings and knowledge-graph data. | Self-hosted, Server | The repository is public and identifies RAPTOR as open source under the Apache-2.0 license. The Aigle 0.4 documentation labels the release a community beta. A development evaluation API is also described for testing without setting up infrastructure. |
| Agenite | Define typed tools with JSON Schema validation and error handling. | CLI | |
| Veldra | Create an agent with a policy, tools, skills, knowledge base and reasoning method. | Self-hosted | The documented deployment is self-hosted and uses a sign-in gate after a first-run administrator account is created. |
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- DHT-AI-Studio/RAPTOR · Official homepage title · retained in its source language