Llamatik Server
Llamatik Server is a self-hosted HTTP backend that exposes the complete Llamatik library API over HTTP. Deploy it on any JVM machine and point your Llamatik apps — or any HTTP client — at it.
All endpoints mirror the bridge surface of com.llamatik:core 2.2.0:
| Bridge | Route prefix | Capabilities |
|---|---|---|
LlamaBridge | /v1/generation | Text generation, streaming SSE, JSON output, KV sessions, MTP speculative decoding |
LlamaBridge | /v1/embeddings | Embedding vectors for RAG and search |
WhisperBridge | /v1/whisper | Speech-to-text, per-segment timestamps, speaker turns |
StableDiffusionBridge | /v1/stablediffusion | Text-to-image, image-to-image, text-to-video (Wan 2.1/2.2) |
MultimodalBridge | /v1/multimodal | Vision-language streaming analysis |
Typical use cases
- Shared inference — one powerful machine running models; many apps calling the same server
- Hybrid mode — apps run on-device by default and fall back to this server when the device is too slow or the model too large
- Heavy workloads — GPU-backed multimodal or video generation on a dedicated machine, consumed by any client over the network
Using it from the Llamatik app
The Llamatik app can discover and connect to a server from Settings → Connect to Server, stream responses from it, and show its status in the chat toolbar. See Connecting to a server.
Connections are currently unencrypted (plain HTTP). Do not send sensitive data over untrusted networks. Encrypted connections are planned; see Deployment for TLS via a reverse proxy.
Quick links
- Getting started — run the server, load a model, make your first request
- Configuration — env vars, authentication, path security
- API Reference — complete endpoint documentation
- Deployment — Docker, systemd, HTTPS