OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
RAG & Knowledge Systems

Raft — lumpenspace

Raft is an open-source Python research project that combines fine-tuning with retrieval-augmented generation to create conversational agents modeled on a specific person. Its repository documents source import, retrieval preparation, fine-tuning, and persona serving.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Raft — lumpenspace — OpenViST project profile in RAG & Knowledge Systems.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Raft when exploring a research workflow that combines a fine-tuned persona model with retrieval from source material, and when you can work with its documented command-line pipeline.

Who it is for

  • Developers and researchers exploring person-specific conversational agents.
  • Teams with source writing or conversations to use as persona grounding and training material.

Documented uses

  • Import writing or conversations from supported sources, including forums, social networks, publications, feeds, web pages, and PDFs.
  • Prepare source material by chunking it and creating embeddings.
  • Generate and run fine-tuning jobs for a persona.
  • Ask questions about imported content or serve retrieval-augmented persona chat.
  • Have a persona comment on a post using a URL or file as its source.
Before you choose

Requirements & limits

The repository is public, open source, and listed under the MIT license. The README identifies the distribution as raft-ft and the command-line interface as raft.

  • The author describes this repository version as old and abandoned and says they do not guarantee it works.
  • The documented workflow requires handling source import, retrieval preparation, and a fine-tuning stage rather than only configuring a hosted chatbot.
  • The README documents different serving paths for comments: an MLX-converted persona or an OpenAI-compatible endpoint.
  • The listed implementation language is Python.

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

Create and prepare a RAFT persona project

Set up a persona project, import source material, and generate its memory store and training examples.

Before you start

  • Python 3.11 or later.
  • Install the raft-ft distribution; install raft-ft[pdf] for PDF sources or ariadne-x for X/Bluesky sources when needed.
  • For unstructured chat logs that must be structured into conversations, configure the helper LLM through RAFT_LLM_MODEL.

Steps

  1. Install RAFT with the command below.

    Documented command
    pip install -U raft-ft

  2. Create and enter a project directory with raft init my-persona followed by cd my-persona.

  3. Import a source, for example raft fetch url --url , raft fetch substack --blog , or raft fetch lesswrong --user .

  4. Prepare the imported material with raft chunk && raft embed, then generate training examples with raft ft:gen or raft ft:gen --thinking.

Expected outcome

The project contains imported source material, a chunked and embedded memory store when grounding documents are present, and generated training examples.

Guide limitations

  • The repository author describes this version as old and abandoned and does not guarantee that it works.
  • Grounding is optional; with conversations alone, prep skips the memory store.
  • A recollection without a verbatim supporting sentence is dropped. --thinking generates reasoning traces after the fact from the target's actual reply.
Documentation 1 ↗
Source-based practical guide

Install RAFT and initialize a persona project

Install the RAFT CLI and create an empty persona project directory.

Before you start

  • Python 3.11 or later.

Steps

  1. Install the distribution with the command below.

    Documented command
    pip install -U raft-ft

  2. Create a project with raft init my-persona.

  3. Enter the project directory with cd my-persona.

Expected outcome

A persona project is initialized in `my-persona`, where subsequent commands can locate the project automatically.

Guide limitations

  • This covers project initialization only; it does not import data, prepare training examples, train a model, or serve a persona.
  • The repository author describes the version as old and abandoned and does not guarantee that it works.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗Guide documentation ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
lumpenspace/raft · Official homepage title · retained in its source language
How we review sourcesSuggest a correction