Skip to content

Quickstart (Ollama)

What success looks like

By the end of this tutorial, Ollama should be reachable, ~/.loopforge/config.toml should be valid, and one workspace artifact should be created by loopforge agent run.

This tutorial runs LoopForge locally using Ollama’s OpenAI-compatible endpoint.

Prerequisites

  • Ollama is installed and running.
  • You have at least one chat model available (example: qwen3:4b, llama3.2, etc.). (Embedding-only models won’t work.)

Check your local models:

1
ollama list

LoopForge defaults to providers.ollama.default_model = "llama3.2" in ~/.loopforge/config.toml.

If you don’t have llama3.2 installed, pick one of these:

1) Pull it:

1
ollama pull llama3.2

2) Or switch LoopForge to a model you already have (example: qwen3:4b):

1
2
[providers.ollama]
default_model = "qwen3:4b"

1) Start Ollama

1
ollama serve

2) Initialize LoopForge

This creates: - ~/.loopforge/config.toml (provider config + routing) - ~/.loopforge/loopforge.db (SQLite memory)

1
loopforge init

3) Run your first agent session

Pick a workspace directory (tools are sandboxed to this root):

1
2
3
mkdir -p loopforge-work
loopforge agent run --workspace loopforge-work --prompt "Create hello.txt with the word hi"
cat loopforge-work/hello.txt

1
2
3
mkdir loopforge-work
loopforge agent run --workspace loopforge-work --prompt "Create hello.txt with the word hi"
Get-Content .\loopforge-work\hello.txt

LoopForge prints the final assistant output, and persists a stable session_id under loopforge-work/.loopforge/session_id.

4) Re-run in the same workspace (optional)

1
loopforge agent run --workspace loopforge-work --prompt "Now append a newline + bye to hello.txt"

Next steps

  • Use the harness for long tasks: see “Long Task With Harness”.
  • Switch providers/models: see “Providers & Routing”.