LLMStack

What is LLMStack?

LLMStack is a no-code platform for building generative AI apps, chatbots and agents that chain several language models together and draw on your own documents. It is the self-hostable engine behind Promptly, a hosted service run by MakerDojo, Inc., the company named in the llmstack.ai copyright line.

You install it on your own machine or server, open a visual editor in the browser, and connect processors from providers such as OpenAI, Anthropic, Cohere, Stability AI, ElevenLabs and Google. Imported files and web pages are split, embedded and stored in a bundled Weaviate vector database. Finished apps can run as web chat, an HTTP API, or bots in Slack and Discord.

Maintenance has slowed sharply. The latest GitHub release is v0.2.6 from 5 November 2024, which added voice agents through the OpenAI Realtime API and Twilio, and the newest commit on the main branch is dated 11 December 2024. The repository shows no archive notice, but nearly two years without code changes means you should treat it as dormant.

Key Features

Model chaining is the core of LLMStack, letting one app pass output from a text model into image, speech or search steps without code.

• Multi-model chaining: You link processors from different providers in a single flow, for example a GPT model that writes a script and an ElevenLabs voice that reads it. The source tree includes processor folders for Anthropic, Mistral, Meta, Google and Stability AI.
• Data import and RAG: You can upload CSV, TXT, PDF, DOCX and PPTX files or pull content from websites, Google Drive and Notion. LLMStack chunks and vectorises the material automatically, so chatbots can answer questions grounded in your own documents.
• Agents and voice: The builder supports agents that pick tools to complete a task, and release v0.2.6 added voice agents. Those use the OpenAI Realtime API with Twilio for phone calls, plus noise reduction added in a later commit.
• Slack and Discord bots: Any app can be triggered from Slack or Discord through settings in the app editor. Teams can therefore put an internal knowledge bot inside the chat tool they already use daily.
• HTTP API access: Each app exposes an API endpoint, so developers can call a no-code flow from their own product. This mixes visual building with conventional software, which suits small teams with one developer.
• Multi-tenant workspaces: Organisations, users and permission levels are built in, with viewer and collaborator roles on each app. Administrators can also set default provider keys for everyone in a configuration file.

Pros

Control over hosting is LLMStack’s strongest argument, because you can run the whole stack on your own hardware and pay only for the model calls you make.

• Free self-hosting: Installing with pip or Docker costs nothing beyond your server and provider keys. The Elastic License 2.0 allows internal and commercial use, which suits teams avoiding per-seat fees.
• Data stays in-house: Documents, vectors and chat logs sit in your own Postgres and Weaviate containers. This helps teams with confidentiality rules that prevent sending files to a third-party builder.
• Wide provider choice: Processors cover text, image, speech and video providers instead of a single vendor. You can swap models or combine them to balance quality, speed and cost.
• No-code with an API: Non-developers can assemble flows visually while developers call them over HTTP. Both groups work on the same app, which shortens the path from prototype to internal tool.
• Useful templates: The project ships example apps and sheet templates for tasks such as content generation, research and sales prospecting. These give new users a working starting point instead of a blank canvas.

Cons

Stalled development is the serious risk, since no commit has landed on the main branch since December 2024 and model APIs keep changing.

• Little recent maintenance: The last release, v0.2.6, dates from November 2024. Newer models, API changes and security patches may not be reflected, so you would need to patch the code yourself.
• Not strictly open source: Directories call it open source, but the Elastic License 2.0 bars offering it to others as a hosted service. Agencies planning to resell it should read the licence carefully first.
• Setup needs technical skills: Self-hosting requires Python 3.10 or later, Docker and, on Windows, WSL2. The default admin password is published in the README, so it must be changed straight after install.
• Bring your own keys: Every model call uses your own OpenAI, Anthropic or other provider account. Costs can climb with busy chatbots, and you manage billing across several vendors.
• Thin community support: Documentation lives on the Promptly docs site, and help mainly comes through GitHub issues and Discord. With slow maintenance, answers to new problems may never arrive.

Who is Using LLMStack?

Technical teams that want a private, self-hosted AI app builder are LLMStack’s natural audience, especially for internal chatbots and prototypes.

• Internal IT teams: They host it on company servers to build document chatbots. Staff can query policies and manuals without data leaving the network.
• Developers and makers: They prototype multi-model apps visually and call them through the API. This saves writing orchestration code for a proof of concept.
• Sales and research teams: They use templates for prospect research and content drafting. Integrations with services such as Apollo appear in the processor list.
• Community managers: They connect knowledge bots to Discord or Slack servers. Members get answers drawn from documentation without waiting for a moderator.

Pricing

LLMStack itself costs nothing to self-host, while the hosted Promptly cloud runs from a free tier to $99.99 a month (official sites, October 2026).

Directories disagree: Futurepedia labels LLMStack as contact for pricing, while There’s An AI For That lists it as fully free. The self-hosted edition is free, and the paid tiers apply only to the Promptly cloud.

Disclaimer: the Promptly site still carried a 2024 copyright line, so confirm the plans are current and still sold before relying on them.

What Makes LLMStack Unique?

A single visual builder that you can host yourself, yet that also chains text, image, voice and video providers, is the concrete differentiator. Many no-code chatbot builders are cloud-only, while many self-hosted frameworks expect you to write Python. LLMStack sits between them, with a browser editor on top of a Django and React codebase you control.

The other unusual piece is the link to Promptly, which lets a team prototype in the hosted cloud and move to its own servers later on the same engine. That path matters less now that the open repository has gone quiet, but the code remains available to fork.

Compatibilities and Integrations

Fifteen-plus provider integrations ship in the processor library, from OpenAI and Anthropic to Twilio, Pinecone and Qdrant.

• Install targets: It runs on Linux and macOS with Python 3.10 or later, and on Windows through WSL2. Docker is needed for background jobs and the bundled services.
• Model providers: Processor folders cover OpenAI, Azure, Anthropic, Cohere, Google, Mistral, Meta, Stability AI, ElevenLabs and HeyGen. You add your keys on the settings page.
• Data sources and stores: Imports include files, websites, Google Drive and Notion. Weaviate is bundled, with Pinecone, Qdrant and SingleStore processors also present.
• Channels: Apps can be embedded on websites, called over an HTTP API, or run as Slack and Discord bots. Twilio support adds phone-based voice agents.
• Promptly cloud: The hosted service mentions Zapier among its integrations. That claim applies to Promptly rather than the self-hosted repository.

LLMStack Tutorials

A working local instance needs Python, Docker and about ten minutes, after which the browser editor handles the rest.

1. Install Python 3.10 or later and Docker, using WSL2 if you are on Windows.
2. Run pip install llmstack, then start the server with the llmstack command.
3. Open localhost:3000, log in with the default admin account and change the password at once.
4. Add your OpenAI, Anthropic or other provider keys on the settings page.
5. Create a data source by uploading files or adding a website, Google Drive or Notion link.
6. Build an app from a template or blank flow, test it, then publish it to the web, API, Slack or Discord.

How We Rated It

Strong features and free self-hosting lift LLMStack to 3.8, but stalled maintenance since late 2024 drags down support and long-term reliability.

Summary

LLMStack suits technical teams that want a free, self-hosted way to build multi-model chatbots and agents over their own documents. Its main limitation is maintenance, because the repository has seen no commits since December 2024 and model APIs move quickly. The Elastic License 2.0 also rules out reselling it as a hosted service. For an internal prototype it remains a capable starting point, but production use means owning updates yourself. Pin the Docker image version and test each provider connection before you rely on it.