Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarOllama Blog→ original

Ollama launch runs Claude Code and Codex on local models without config files

Connecting Claude Code or Codex to a local model usually means setting environment variables and editing config files. The ollama launch command removes that hassle: you pick a model, and Ollama starts Claude Code, OpenCode, or Codex for you. Running locally with a 64,000-token context requires about 23 GB of VRAM; otherwise, cloud models are available. This makes it possible to test agentic coding on your own machine with a single command.

AI-processed from Ollama Blog; edited by Hamidun News
Ollama launch runs Claude Code and Codex on local models without config files
Source: Ollama Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceOllama Blog ↗
Editor-in-Chief
Locally~23 GB VRAM at a 64,000-token context
Key metric #2Cloud models provide the full context, and limits are generous even on the free tier
Key metric #3The coding session window has been extended to 5 hours, and limits have increased

Coding agents such as Claude Code, OpenCode, and Codex have long since stopped being toys, but running them on your own model is still often tedious work. You have to find the right environment variables, set the address of the local server, and figure out the config format for each tool. Ollama has decided to eliminate this routine and has added the ollama launch command, which handles the setup for you.

What ollama launch does

The idea is simple: you run one command in the terminal, Ollama shows a list of models and integrations, you choose what you want, and the chosen tool starts already connected. According to the developers, neither environment variables nor configuration files need to be edited by hand. Popular coding tools are supported: Claude Code, OpenCode, and Codex.

This works with two types of models. The first option is local models that run on your graphics card. The second is cloud models that Ollama hosts itself. For a developer, the difference comes down to selecting an item from a menu rather than rewriting settings for each backend.

If you only want to prepare the tool without launching it right away, there is a configuration flag. For example, the command ollama launch opencode --config will configure OpenCode but will not open it. This is useful when the environment is being set up in advance, say, for a team or for a deployment script.

How much hardware and context you need

The main practical figure from the announcement is the memory requirement. Running the recommended coding models locally requires about 23 GB of VRAM with a 64,000-token context. That is the level of top consumer graphics cards or a workstation, not an ordinary laptop.

The announcement addresses context separately: coding tools work best with the full context length. An agent reads many files, the history of edits, and command output, so a short window fills up quickly and the agent starts losing the thread. Ollama advises setting the context length to at least 64,000 tokens in the settings. If you leave the default value, the agent may behave noticeably worse, and the culprit will be the truncated window, not the model.

For the reader, the takeaway is this: if local running doesn't work, first check the context length, and only then blame the model.

The cloud option and limits

For those who don't have 23 GB of VRAM, Ollama offers cloud models. They work with the full context, and, as the company promises, limits remain generous even on the free tier. Along with the update, Ollama has increased quotas and extended the coding session window to five hours. The announcement does not give specific figures for requests and tokens, so it is best to check the real ceiling of the free tier yourself on your own tasks.

What this means in practice

For teams considering local models because of code privacy or API bills, the barrier to entry has noticeably dropped. Previously, comparing how Claude Code behaves on a local model versus a cloud model meant spending an evening on setup. Now it takes a couple of commands.

There are limitations, too. Agent quality depends on the model, not on a convenient launcher: a local model that fits in 23 GB is not guaranteed to handle the same tasks as flagship cloud models. Ollama's cloud option sends code to third-party servers, which for many is the very reason to move away from the cloud. In addition, the announcement does not say which models and tools besides those named are supported, so the list should be checked in the documentation.

A reasonable approach: use ollama launch as a quick test bench. Run the same tool on a local model and a cloud model, compare the results on real tasks from your own repository, and only then decide whether to move your workflow to local models.

Story

Key Questions Answered

What ollama launch does?

The idea is simple: you run one command in the terminal, Ollama shows a list of models and integrations, you choose what you want, and the chosen tool starts already connected. According to the developers, neither environment variables nor configuration files need to be edited by hand. Popular coding tools are supported: Claude Code, OpenCode, and Codex. This works with two types of models. The first option is local models that run on your graphics card. The second is cloud models that Ollama hosts itself.

How much hardware and context you need?

The main practical figure from the announcement is the memory requirement. Running the recommended coding models locally requires about 23 GB of VRAM with a 64,000-token context. That is the level of top consumer graphics cards or a workstation, not an ordinary laptop. The announcement addresses context separately: coding tools work best with the full context length. An agent reads many files, the history of edits, and command output, so a short window fills up quickly and the agent starts losing the thread.

What are the key details on The cloud option and limits?

For those who don't have 23 GB of VRAM, Ollama offers cloud models. They work with the full context, and, as the company promises, limits remain generous even on the free tier. Along with the update, Ollama has increased quotas and extended the coding session window to five hours. The announcement does not give specific figures for requests and tokens, so it is best to check the real ceiling of the free tier yourself on your own tasks.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…