What it does
Core capabilities at a glance
- Cuda
- Deepseek
- Deepseek V4
- Gguf
- GLM
- LLM Inference
- Local LLM
- Mixture OF Experts
Deep dive
The full breakdown - performance, comparisons, and setup
Xolotl
Xolotl is an agent framework - Xolotl (formerly Strata-DSV4.1): run DeepSeek-V4.1-Flash, all 552B, as a local coding agent on one gaming PC (RTX 4090 + 90 GB RAM) at up to ~27 tok/s, 512K context, October 1017 builds included. Plus GLM-5.3-Flash and Qwen3.8. MoE experts tiered across VRAM, pinned RAM and NVMe; OpenAI + Anthropic API.
Overview
DeepSeek-V4.1-Flash - all 552 billion parameters - on one gaming PC, at 27 tok/s. Plus GLM-5.3-Flash, Qwen3.8-Flash-Next and Qwen3.8-27B: one engine, one OpenAI / Anthropic-compatible server. Built on Strata and heavily extended · Linux · NVIDIA · MIT
One RTX 4090 (24 GB) · Ryzen 7 5700X · 90 GB DDR4 · one NVMe drive. Every number is measured on that machine, through the real server, while it solved real tasks (21 HumanEval problems, 10 tool-call tasks, 3 agentic bug-fix loops), greedy, thinking on with a 32K-token thinking budget.
- A 552B frontier model as a local coding agent. Point Claude Code, Codex, opencode or Open WebUI at it - tool calls, streaming, prefix reuse across agent turns, a thinking budget. - No server farm. The weights (152 GiB for DeepSeek-V4.1) live across VRAM, pinned RAM and NVMe; the engine carries up the experts each token needs. - Fast where it counts. The Qwen models run at 100-130 tok/s; an agent's small follow-up turn on DeepSeek-V4.1 costs ~8 s on a 6K-16K context, and the engine learns the conversation's experts between turns.
Xolotl is open-source, written primarily in C++, with 419 GitHub stars under the MIT license. It was last updated on 2026-10-30.
Key capabilities
From the project's documentation:
- A 552B frontier model as a local coding agent. Point Claude Code, Codex, opencode or Open WebUI at it - tool
- No server farm. The weights (152 GiB for DeepSeek-V4.1) live across VRAM, pinned RAM and NVMe; the engine
- Fast where it counts. The Qwen models run at 100-130 tok/s; an agent's small follow-up turn on DeepSeek-V4.1
- Deep sleep for MiniMax-M3.1 and DeepSeek-V4.1 (#75, NektoDron): opt-in "deep_sleep": "auto" | "native" |
- Qwen3.6 vision, shared prefix cache and batched sampled MTP (#76, NektoDron): text and image prefixes are reused
- Qwen3.6 resident prefix reuse (#80, NektoDron, opt-in --resident-prefix-reuse): when the immutable host pages
How it fits a local-AI stack
Xolotl runs on your own hardware, so pair it with a model and a GPU sized to your needs. Use the VRAM calculator to pick a model that fits your card, and see what you can run for hardware guidance. Related agent frameworks in the directory:
Sources
- Source code & docs: Indras-Mirror/Xolotl
- Official website: https://github.com/Niko1221/Strata
Stats from GitHub, 2026-10-30.
Frequently asked
Quick answers to common questions
What is Xolotl?
Xolotl is a agent-framework tool for local AI workloads. Xolotl (formerly Strata-DSV4.1): run DeepSeek-V4.1-Flash, all 552B, as a local coding agent on one gaming PC (RTX 4090 + 90 GB RAM) at up to ~27 tok/s, 512K co…
Is Xolotl free and open source?
Yes, Xolotl has 419 GitHub stars and is licensed under MIT. You can self-host it for free on linux.
What platforms does Xolotl support?
Xolotl runs on linux.
What hardware do I need for Xolotl?
The hardware requirements depend on which models you run. Check our hardware directory for compatible GPUs and systems. Xolotl has 419 GitHub stars and an active community.
Does Xolotl support GPU acceleration?
Xolotl supports GPU acceleration via CUDA, Metal, or Vulkan depending on your platform. For the best performance, pair it with an NVIDIA RTX 4090 or 5090.
What are the best alternatives to Xolotl?
Popular alternatives include other agent-framework tools in our directory. Browse our full collection at /tool for comparisons, community reviews, and benchmark data to find the right fit for your workflow.
How much does Xolotl cost?
Xolotl is free-open-source. It is completely free and open source to self-host.
Pairs well with
Complementary tools, models, and hardware