Skip to content
NeoStack
Esc
↑↓navigate↵open⌘Jpreview
On this page

Ollama

Use Ollama local models with NeoStack AI.

Ollama is the easiest local LLM option because NeoStack AI has a built-in Ollama (Local) provider.

Setup

  1. Install Ollama from ollama.com.
  2. Pull a model:
ollama pull qwen2.5-coder
  1. Make sure Ollama is running.
  2. Open Tools > Agent Chat.
  3. Open Settings > Chat & Agents > Chat Providers.
  4. Add or enable Ollama (Local).
  5. Keep the default base URL:
http://localhost:11434/v1
  1. Start a chat with Local & BYOK Chat and pick your Ollama model.

Models

NeoStack AI discovers installed Ollama models from Ollama’s local model list. If a model does not appear, pull it first:

ollama pull llama3.2
ollama pull deepseek-r1
ollama pull qwen2.5-coder

Requirements

  • Ollama 0.8.0 or newer. Older versions cannot stream responses when tools are attached, and NeoStack AI always attaches its editor tools. Run ollama -v and upgrade if needed.
  • A tool-capable model. NeoStack AI sends its editor tools with every chat request, and Ollama rejects that with an HTTP 400 (“does not support tools”) for models whose template has no tool support (e.g. gemma3, deepseek-r1). Pick a model listed with the tools tag on ollama.com/search, such as qwen2.5-coder, qwen3, llama3.1, llama3.2, or mistral.
  • A larger context window. Ollama’s default context is small (4096 tokens), and the editor tool definitions plus your conversation easily exceed it — overflow is silently truncated, which makes the model ignore instructions or produce broken tool calls. Raise it before starting Ollama:
# Windows (PowerShell): setx OLLAMA_CONTEXT_LENGTH 16384, then restart Ollama
export OLLAMA_CONTEXT_LENGTH=16384

Troubleshooting

  • “HTTP transport failed (ConnectionError)” — either Ollama is not running at the configured base URL, or (on plugin versions before the activity-timeout fix) the model took longer than 30 seconds to load / process the prompt before sending its first byte. Warm the model first (ollama run <model> "hi"), prefer a smaller model, or update the plugin.
  • “… does not support tools” (HTTP 400) — the selected model has no tool template; switch to a tool-capable model (see above).
  • Models appear in the picker but every chat fails — check the base URL. It must be the OpenAI-compatible root (http://localhost:11434/v1); newer plugin builds normalize a bare http://localhost:11434 automatically.

Notes

  • Ollama does not need an API key for normal local use.
  • NeoStack AI uses Ollama’s OpenAI-compatible chat endpoint for chat.
  • Model discovery uses Ollama’s native /api/tags endpoint so installed local models show up cleanly.

Was this page helpful?