# Llama Stack Local Server for development purposes

> Run model for the first time to download/install it into the local Ollama server.

- Skill: `tools-only/llama-stack-local-server-for-development-purposes` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add tools-only/llama-stack-local-server-for-development-purposes`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tools-only/llama-stack-local-server-for-development-purposes/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: tools-only (https://skillmd.com/u/tools-only)
- Updated: 2026-09-29
- Page: https://skillmd.com/skills/tools-only/llama-stack-local-server-for-development-purposes

---

# Llama Stack Local Server for development purposes

## Run Llama Stack

Install [Ollama](https://ollama.com/download) and make sure it is running.

Run model for the first time to download/install it into the local Ollama server.

```sh
ollama run granite3.2:2b
```

Tested models: 
* `granite3.2:latest` - 8B model
* `granite3.2:2b` - 2B model
* `granite3.1-dense:2b`
* `granite3.1-dense:8b`
* `llama3.2:latest`

Create empty `~/.llama` directory first, if it doesn't exist on your filesystem. It is used to persist LlamaStack platform data (configuration, etc).

Ollama model must be running during the LlamaStack startup, so you have to run it shortly before each LlamaStack start, or run it with `-keepalive` in another terminal.

```sh
ollama run granite3.2:2b --keepalive 60m 
```

Start [LlamaStack Starter distribution](https://llama-stack.readthedocs.io/en/latest/distributions/self_hosted_distro/starter.html) container.

Version of the LlamaStack server distribution must be the same as version of the [LlamaStack client used by the UI Agent, see](libs/3rdparty/python/llama-stack-client-constraints.txt)!

```sh
podman run -it --rm \
  -p 5001:5001 \
  -v ~/.llama:/root/.llama:z \
  llamastack/distribution-starter:0.2.20 \
  --port 5001 \
  --env INFERENCE_MODEL="granite3.2:2b" \
  --env OLLAMA_URL=http://host.containers.internal:11434
```
**Notes:** 
* `:z` qualifier on `/root/.llama` volume mount in necessary for Fedora Linux
* use `--env INFERENCE_MODEL="granite3.2:2b" \` with another model to add it into your LlamaStack instance, it remembers models added before and all are available then

Output:

```sh
INFO 2025-03-14 08:10:46,174 llama_stack.providers.remote.inference.ollama.ollama:74: checking connectivity to Ollama at `http://host.containers.internal:11434`...
INFO 2025-03-14 08:10:46,453 httpx:1740: HTTP Request: GET http://host.containers.internal:11434/api/ps "HTTP/1.1 200 OK"
...
INFO:     Started server process [1]
INFO:     Waiting for application startup.
INFO:     ASGI 'lifespan' protocol appears unsupported.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://['::', '0.0.0.0']:5001 (Press CTRL+C to quit)
```


You can use [**`run-llamastack-ollama.sh`**](run-llamastack-ollama.sh) script found in this directory, which performs described steps to run LLamaStack.

You can export `INFERENCE_MODEL` environment variable before running it, to select different LLM model to be used. This environment variable is 
shared with other tools in this project, like evaluation framework.