Models
Bring the runtime you trust.
TuxNest does not depend on a publisher-hosted inference service. It can discover local Ollama models, tunnel to an Ollama host you own, or use an explicitly configured OpenAI-compatible API.
Provider paths
Same editor. Different compute.
| Runtime | Where inference runs | What TuxNest does | Best for |
|---|---|---|---|
| Local Ollama | Your development machine | Discovers installed models; may start an existing supported local Ollama installation. | Privacy-first local work |
| Remote Ollama | Your SSH-accessible GPU host | Creates an SSH tunnel, discovers that host's installed models and surfaces connection/GPU status. | Larger models |
| OpenAI-compatible | Your configured endpoint | Sends the selected context to the explicit base URL you configure. | vLLM · LM Studio · llama.cpp |
Effort
Context is a budget, not a promise.
4K context ceiling
Up to 4096 context tokens and 2048 output tokens for Ollama responses, subject to lower configured/model limits.
8K context ceiling
Up to 8192 context tokens and 4096 output tokens, bounded by actual model configuration.
16K context ceiling
Up to 16384 context tokens and 8192 output tokens when the selected runtime supports it.
Configured maximum
Uses the configured maximum and can remove the normal agent-round cap; it is still finite hardware.
TuxNest hybrid compute
Your GPU. Your AI. Your Control. Anywhere.
TuxNest connects a reviewable VS Code coding workflow with selected local models, SSH-tunneled models on user-controlled GPU workstations and explicitly configured compatible endpoints.