Skip to main content

Connect Ollama

Ollama lets you run AI models directly on your computer. No API key, no cloud account, no usage costs. The model runs entirely on your machine.

This is the recommended way to run Manasvi locally.

If you prefer a cloud model instead, see Connect Claude or Configure a Model.


What is Ollama?

Ollama is a tool that downloads and runs AI language models on your computer. Once you've pulled a model, it works offline — your data never leaves your machine.

Supported models include Llama 3, Mistral, Qwen, Phi, Gemma, and many others. All free to download and use.


Step 1 — Install Ollama

Download and install Ollama from ollama.com. It's available for macOS, Linux, and Windows.

After installing, start the Ollama server:

ollama serve

You should see output like:

Ollama is running on http://localhost:11434

Leave this running in the background while you use Manasvi.


Step 2 — Pull a model

# Llama 3.2 (3B) — recommended for most laptops, fast
ollama pull llama3.2

# Llama 3.2 (8B) — better quality, needs more RAM
ollama pull llama3.2:8b

# Mistral — fast, good general performance
ollama pull mistral

# Qwen 2.5 — strong instruction following
ollama pull qwen2.5

Which model should I choose?

ModelSizeRAM neededBest for
llama3.2 (3B)~2GB4GBFast responses, lower-spec machines
llama3.2:8b~5GB8GBBetter quality, good for most tasks
mistral (7B)~4GB8GBFast, general purpose
qwen2.5 (7B)~4GB8GBStrong at following instructions

The download only happens once. After that, the model is stored locally.


Step 3 — Configure Manasvi

pnpm manasvi models add ollama

This prompts you for the model name and writes the settings automatically.

If you're doing this during first-time setup, use:

pnpm manasvi onboard

And select Ollama (local) when prompted.

Manually

Add these to your .env.local file:

MODEL_ADAPTER_MODE=ollama
OLLAMA_BASE_URL=http://localhost:11434/v1
PLANNER_MODEL=llama3.2

Then restart Manasvi:

pnpm manasvi restart

Step 4 — Verify the connection

pnpm manasvi models test

This sends a test request to Ollama and reports success or the specific error.

pnpm manasvi doctor

The doctor command checks whether Ollama is reachable and reports its status in the Model Backend section.


Switch between models

To try a different model without re-running onboard:

# Pull the model first
ollama pull qwen2.5

# Then update the config
pnpm manasvi models use ollama

It will prompt you for the model name.

Or edit .env.local directly:

PLANNER_MODEL=qwen2.5

And restart:

pnpm manasvi restart

Troubleshooting

"Ollama is not running"

Start Ollama first:

ollama serve

Check it's working:

curl http://localhost:11434/api/tags

You should see a JSON list of your pulled models.

"Model not found"

Make sure you've pulled the model you specified:

ollama list # see what's installed
ollama pull llama3.2 # pull if missing

Slow responses

Local models are slower than cloud APIs, especially on first load (the model needs to load into memory). After the first request, subsequent ones are faster.

To check if Ollama is overloaded:

ollama ps

Out of memory

Try a smaller model:

ollama pull llama3.2 # 3B version uses less RAM than 8B

Or close other memory-heavy applications while Manasvi is running.


Using a different Ollama port

If Ollama is running on a non-default port, update your .env.local:

OLLAMA_BASE_URL=http://localhost:11435/v1

Or run pnpm manasvi models add ollama and enter the custom URL when prompted.