Local runtime setup

Use Ollama with Kilo Code for Local AI Coding

Ollama is a practical choice when you want a scriptable local model runtime with a simple model library and a service that can stay running in the background. Kilo uses the local provider you configure, so you can keep model inference on hardware you control.

Install Kilo Code

Download Ollama

Best for: Developers who prefer a lightweight, command-line-friendly runtime
Connection: Kilo local Ollama provider
Model management: Pull and manage models with Ollama

Connect Ollama to Kilo

1
Install and start Ollama
Install Ollama from its official site and confirm that the local service is running before opening Kilo provider settings.
2
Pull a model that fits your hardware
Choose the exact model artifact and quantization deliberately. Leave memory headroom for your IDE, repository index, tools, and the context window.
3
Select Ollama in Kilo
Open Kilo provider settings, choose Ollama, and select the model exposed by your local runtime. Keep the runtime on its local address unless you intentionally operate a secured remote endpoint.
4
Validate a small coding task
Start with a bounded request that reads one file and proposes a small edit. Confirm model responses and tool calls before using longer agent workflows.

Before you connect

Local model quality varies. A model that answers coding questions may still struggle with reliable tool use or long agent loops.
Long context windows consume additional memory, and runtime defaults can differ from a model’s advertised maximum context.
Local inference removes hosted token charges, but hardware, power, setup time, and maintenance still have costs.
Ollama Cloud is a hosted service and is different from running the Ollama runtime on your own machine.

Troubleshooting

Kilo cannot reach Ollama: Confirm that the Ollama service is running and that Kilo is configured for the same local endpoint. Check local firewall or proxy rules before exposing the service beyond localhost.
The model is missing: Confirm that the model finished downloading and is listed by Ollama, then reopen or refresh model selection in Kilo.
Prompts are truncated or the agent loses context: Check the runtime context configuration and available memory. Reduce repository context or select a model and quantization with enough memory headroom.
Tool calls are unreliable: Use a model with demonstrated instruction-following and tool-use ability, and validate it on a small task before increasing autonomy.

Ollama and Kilo FAQ

Can I use Ollama with Kilo Code in VS Code, JetBrains, and the CLI?

Yes. Kilo supports Ollama as a local model provider across its local coding clients: the VS Code extension, JetBrains plugin, and CLI.

Does using Ollama make every Kilo feature offline?

No. Ollama can keep model inference local, but sign-in, updates, remote MCP servers, telemetry choices, Cloud Agents, hosted code review, and other networked features have separate data paths. An air-gapped workflow requires configuring or disabling each networked dependency.

Is Ollama inference free?

Ollama avoids a hosted per-token bill when it runs on your hardware. You still supply the computer, memory, storage, electricity, setup, and maintenance.

Choose the right local coding model

Compare evidence-backed local model profiles by runtime, quantization, memory, and hardware tier before downloading a large artifact.

View local model recommendations Read Kilo local model docs