Which Ollama models should I use?

A coder instruct model for chat and edits, a small base coder model for completion, and an embedding model for the index. The supported-models page in the docs lists what works on what hardware.

Does twinny need Ollama running on the same machine?

No. Point a provider at any Ollama address on your network, pair a second computer through Devices, or run a team gateway in front of it.

Does anything leave my machine?

Only the requests to the Ollama address you configured. No telemetry, no account, no update check.

twinny with

VS Code, with Ollama doing the work.

Ollama is the local server twinny supports best. It runs as a background service, pulls models with one command, and has native endpoints for completion and embeddings as well as an OpenAI-compatible chat API. twinny finds it on its usual port the first time it starts and sets up a provider for every job it has a model for.

install for VS Code Ollama in the docs live demo

three steps

01
Install Ollama from ollama.com and pull a chat model, a small completion model, and an embedding model if you want the workspace index.
02
Install twinny from the Marketplace. On first start it finds Ollama on port 11434 and lists its models; pick one per job.
03
Type. Completions appear as ghost text; Ctrl+I edits in place; the sidebar has chat, review and the index.

Native endpoints

Autocomplete uses /api/generate with a fill-in-the-middle template chosen from the model name; embeddings use /api/embed; chat uses the /v1 chat API. Model names are Ollama tags, exactly as ollama list prints them.

Another machine, same Ollama

Pair two of your computers with a code and use the GPU on the other one over an encrypted peer-to-peer link. No account, no relay.

Tuned for small models

Completion is cut off at a sensible end and works well with a 1.5B to 7B coder model; chat and edits are happy on 7B.

One gateway for the team

twinny-server sits on the machine with the models and gives every developer a key, usage per person, policy and an admin page. Free for 5 developers, $6 a seat after that, 30-day trial.

questions

Which Ollama models should I use?: A coder instruct model for chat and edits, a small base coder model for completion, and an embedding model for the index. The supported-models page in the docs lists what works on what hardware.
Does twinny need Ollama running on the same machine?: No. Point a provider at any Ollama address on your network, pair a second computer through Devices, or run a team gateway in front of it.
Does anything leave my machine?: Only the requests to the Ollama address you configured. No telemetry, no account, no update check.

Everything else is on the home page: features, the team gateway, security and pricing.