LoginSign Up
← Back to all sparks
A

AnythingLLM

AI-ML-PLATFORMS
Velocity3.1

All-in-one AI application with RAG and agents, desktop or self-hosted

anythingllm.com ↗

AnythingLLM's Docker image can now run models itself, no Ollama or LM Studio required.

local llmself-hosteddockerai agentsopen source
◆Current state
AnythingLLM 1.19.0 adds an opt-in built-in LLM to the Docker image: AMD's open-source Lemonade server running llama.cpp, with CUDA, Vulkan or CPU picked automatically on amd64 and arm64. When enabled it becomes the default provider at onboarding, recommends three models for the hardware, and pulls any GGUF from Hugging Face. The release also adds UI color palettes and a long list of fixes, including a request-body cap against unauthenticated embed DoS and formula neutralization in CSV exports.
◆Where it's heading
After two fix-only releases, the product is closing the gap between Desktop and self-hosted Docker: the bundled inference engine that Desktop already had now ships in the container. The pattern is to absorb dependencies users previously ran beside it, from inference here to transcription and image generation earlier.
◆Prediction
Expect follow-up releases tuning the built-in engine (model recommendations, GPU setup, Model Router use) and expanding the new palette system toward layout customization, which the notes say is planned.

◆Recent moves

  1. 2d ago

    AnythingLLM v1.19.0

    ⚡ SPARK

    Ends the two-release stabilization stretch with a structural change: the Docker build now bundles its own inference engine, as Desktop already did. Self-hosted deployments no longer need a separate model server to work out of the box.

    View source ↗
  2. 15d ago

    Fixes for agent tool calls, Outlook replies and scheduled jobs

    1.18.0 is a long list of fixes: agent tools now get required parameters on every provider path, Outlook replies carry attachments, scheduled jobs handle half-hour time zones and stop correctly on Windows, and AstraDB embedding works again. It is a second stabilization release in a row, aimed at the agent and automation features added since 1.15.

    View source ↗
  3. 27d ago

    Fixes for spreadsheet dates, agent socket crashes

    A fix-heavy release: spreadsheet date parsing, Anthropic temperature handling, and agent WebSocket crash guards. Stabilization after the 1.16 feature run.

    View source ↗
  4. 1mo ago

    Vertex AI provider and image generation as an agent skill

    Adds Google Vertex AI and LLMman as providers, a generate-image agent skill, and You.com as default web search. Delivers the agent-side image generation promised in 1.16.0.

    View source ↗
  5. 2mo ago

    Collapsible activity view for long agent runs

    Thinking and tool calls from long agent runs now collapse into one element. UI work aimed squarely at multi-step agent sessions.

    View source ↗
  6. 2mo ago

    Image generation and editing via /img

    Image generation and editing via /img with supported providers, plus folder drag-and-drop uploads. Widens the app beyond text chat.

    View source ↗