OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Memory & RAG

Carrel

Carrel is self-hosted knowledge-base infrastructure for AI agents that parses, indexes and retrieves document content through an API, with optional knowledge graphs.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Carrel — OpenViST project profile in Agent Memory & RAG.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Carrel when an agent needs evidence from a changing collection of internal or project documents and you can operate the self-hosted services and their model endpoints.

Who it is for

  • Teams or developers building AI agents that need retrieval from their own document collections.
  • Operators who want to self-host indexing and retrieval and configure the associated model services.

Documented uses

  • Parse and index files such as PDFs, office documents, images, HTML, Markdown and source code.
  • Keep indexes and enabled knowledge graphs updated as files change.
  • Retrieve relevant source passages, images and related facts for an agent to use in its own answers.
  • Inspect processing tasks, document chunks and graphs in a web console.
Before you choose

Requirements & limits

Carrel is self-hosted and published under the MIT license. The console binds to localhost by default; the project documents host and token settings for LAN access.

  • Deployment requires Python 3.12+ and Docker 25+ with Compose v2.
  • A text-embedding endpoint is required; image descriptions and knowledge-graph construction use separate vision-language and chat model endpoints respectively.
  • The optional local model stack requires an NVIDIA GPU and NVIDIA Container Toolkit; CPU parsing is supported.
  • The documented platform options are Linux, macOS with Docker Desktop, and Windows through WSL2; native Windows deployment is unsupported.
  • The stack uses Qdrant for vectors, OpenSearch for keyword search, Neo4j for graph queries and SQLite for file and task tracking.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Deploy Carrel and index a document folder

Set up the self-hosted services, configure required model endpoints and enable a folder as a knowledge base.

Before you start

  • Python 3.12 or later
  • Docker 25 or later with Compose v2
  • A text-embedding endpoint and a vision-language endpoint for image descriptions
  • For the optional local model stack: NVIDIA GPU and NVIDIA Container Toolkit

Steps

  1. Clone the repository, enter its directory and run ./deploy.sh.

  2. Edit config/knowledge-base.env to set the embedding and vision model URLs, model names and API keys; restart running application services after changes.

  3. Place documents in runtime/mirror//, open the console at http://127.0.0.1:9800 and enable the folder.

  4. Check service health and task progress in the console as scheduled tasks process the files.

Expected outcome

The enabled folder becomes a knowledge base whose documents are processed for retrieval.

Guide limitations

  • First startup downloads parser models.
  • Local model services need a separate setup; Linux hosts without a GPU can connect to model services on another host or a provider.
  • Linux with user systemd can enable services and scheduled tasks automatically; other hosts receive manual startup commands.
Documentation 1 ↗
Source-based practical guide

Connect an agent to Carrel search

Use the documented search API to retrieve source material for an agent workflow.

Before you start

  • A running Carrel retrieval service
  • An agent or HTTP client that can send a JSON search request

Steps

  1. Install the bundled carrel-search skill in the agent, or use the API directly.

  2. Send a JSON POST request to the local search endpoint at http://127.0.0.1:9810/search with a question and top_k value.

  3. If KB_SEARCH_TOKEN is set, include an authorization bearer token; for a remote agent, configure CARREL_SEARCH_BASE_URL and CARREL_SEARCH_TOKEN in its environment.

  4. Use returned source passages and available graph evidence or image information as supporting material for the agent's answer.

Expected outcome

The agent can retrieve passages and, when available, graph evidence and image information from Carrel.

Guide limitations

  • Carrel returns retrieval evidence; the agent generates its own answer.
  • Remote access depends on the corresponding service and agent token configuration.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
FengZhenfei/carrel · Official homepage title · retained in its source language
How we review sourcesSuggest a correction