mintplex-labs/anything-llm
Stop renting your intelligence. Own it with AnythingLLM. Everything you need for a powerful local-first agent experience
How anything-llm is put together
AnythingLLM is a self-hostable full-stack application for turning private documents into an AI chatbot with RAG, agents, and multi-LLM support. It is split into three Node.js services plus a React frontend: a `server` (Express API + Prisma/SQLite persistence and vector-DB integration), a `collector` (Express document-ingestion/parsing service), and a Vite/React `frontend` SPA. Data flows from user uploads through the collector (which parses files/links into documents), into the server where documents are embedded and stored in a vector database and chats are orchestrated across configured LLM providers. The stack is JavaScript throughout, orchestrated as a yarn workspace-style monorepo with Docker deployment support.
Languages
Frameworks
Datastores
Infrastructure
Major components
server
Express backend exposing API endpoints for workspaces, chat, admin, agents, documents, and auth, backed by Prisma models.
collector
Standalone Express service that ingests and parses files, links, and raw text into normalized documents for embedding.
frontend
React/Vite single-page application providing the chat UI, admin settings, onboarding, and workspace management.
server/endpoints
HTTP route handlers grouped by domain (chat, admin, workspaces, agents, embeds, MCP, telegram, etc.).
server/models
Prisma-backed data-access models for users, API keys, workspaces, chats, vectors, invites, and system settings.
server/jobs
Background/scheduled workers for embedding, memory extraction, document sync, and cleanup tasks.
collector/extensions & processors
Pluggable link/file processors (YouTube, web scraping, audio-to-wav, office/PDF parsing) that expand ingestion sources.
open-computer
Auxiliary submodule/CLI tooling packaged alongside the main app for a separate 'open computer' capability.
The team shipped cost tracking and a refreshed brand look while fixing several reliability and accessibility issues.
5 weeksWeek of 2026-08-17
Week of 2026-08-17latest
The team shipped cost tracking and a refreshed brand look while fixing several reliability and accessibility issues.
Week of 2026-08-10
The team released version 1.16.0 with streamed voice playback, image usage metrics, and broader integration support.
Week of 2026-08-03
The team added in-chat image generation and new integrations while overhauling AWS Bedrock support.
Week of 2026-07-27
The team improved agent chat behavior and document handling while adding a new web search option and language.
Week of 2026-07-20
The team focused on chat usability, several new languages, and a security-hardening improvement.
Over the past five weeks the team steadily broadened AnythingLLM's AI provider support (AWS Bedrock, Foundry Local, Gemini) while adding practical everyday features like cost and image-usage tracking, in-chat image generation, and folder uploads. A recurring theme was making agent chats more reliable — letting users stop generation cleanly, saving chat history properly, and preventing crashes. Alongside this, the team invested in accessibility, more languages, streamed voice playback, and a cleaner, more consistent visual design.
Week by week
2026-08-17The team shipped cost tracking and a refreshed brand look while fixing several reliability and accessibility issues.latest6 changes
Feature
Cost tracking
Added the ability to track spending associated with AI usage.
Feature
New AnythingLLM wordmark and logo
Introduced a refreshed brand wordmark and fixed how the logo adapts to light and dark themes.
Refactor
Unified modal design
Standardized pop-up dialogs across the app onto one consistent, reusable component.
Fix
Better scanned-document reading
Black-and-white scanned PDFs are now read via OCR instead of having pages silently dropped.
Fix
Screen-reader announcements
The app now audibly announces when a reply is being generated and when one fails.
Fix
Agent chats saved to history
Agent conversations started via the API are now correctly saved into thread history.
2026-08-10The team released version 1.16.0 with streamed voice playback, image usage metrics, and broader integration support.6 changes
Chore
Version 1.16.0 released
Tagged and published a new major version of the application.
Feature
Streamed voice playback
Text-to-speech now begins playing as it streams and handles long messages more smoothly.
Feature
Image usage metrics
Added tracking of image-related usage within the app.
Feature
GitHub Enterprise support
The GitHub data connector now works with self-hosted GitHub Enterprise URLs.
Fix
Voice stops when switching chats
Voice playback now correctly stops when the user moves to a different conversation thread.
Fix
Corrected Taiwanese Chinese labels
Fixed mistranslated scheduled-job labels in the zh-TW language.
2026-08-03The team added in-chat image generation and new integrations while overhauling AWS Bedrock support.6 changes
Feature
In-chat image generation
Users can now generate images directly using an /img command.
Feature
Gitea data connector
Added support for pulling in data from Gitea repositories.
Feature
Cleaner agent stopping
Stopping generation now properly cancels agent sessions from start to finish.
Refactor
AWS Bedrock overhaul
Reworked and improved support for Amazon's Bedrock AI service.
Fix
Jobs notification indicator
Added a visual indicator on the notification bell for scheduled jobs.
Fix
Milvus document count
Fixed an incorrect document count when using the Milvus vector database.
2026-07-27The team improved agent chat behavior and document handling while adding a new web search option and language.6 changes
Feature
Folder uploads with structure preserved
Users can upload entire folders and keep their original file paths in the document manager.
Feature
You.com web search
Agents can now use You.com as a web search provider.
Fix
Restored stop button during agent sessions
The stop-generation button now works again while an agent session is running.
Fix
More reliable document sync
Automatic re-syncing of watched documents no longer breaks pinned-document de-duplication.
Refactor
Removed workspace profile pictures
Dropped the little-used workspace profile picture feature.
Feature
Croatian language support
Added Croatian as an available interface language.
2026-07-20The team focused on chat usability, several new languages, and a security-hardening improvement.6 changes
Fix
Auto-scroll in chat
Improved automatic scrolling so the conversation follows along as replies arrive.
Fix
Slash commands fill in reliably
Fixed slash commands so they correctly expand into prompt text even with surrounding punctuation or spaces.
Feature
Indonesian language support
Added Indonesian as an available interface language.
Feature
Lao language support
Added Lao as an available interface language.
Fix
Security hardening for agent tools
Applied a defense-in-depth improvement based on a reported security advisory.
Chore
Tests for slash command expansion
Added automated tests covering how slash commands are detected and expanded.
Dependencies and code review
Dependency advisories
Security Watch
No known advisories across 2 scanned dependencies.
No known advisories in the scanned dependencies.
Code review
No concrete code-level issues confirmed.
Get this report every week for your repos.
GitZoid learns each repo, reports what changed, and flags what needs attention. One flat price for the whole team.
$19 a month, flat · First 10 outputs free · No card required