index
Sitemap
Every page on CoderBaba in one place, with a one-line summary of each. Search engines can use the XML version at /sitemap.xml.
Main pages
- Home
CoderBaba, the free, open-source AI coding agent for the command line: build anything from your terminal.
- How CoderBaba works
Install with npm, run tasks in parallel git worktrees, approve every diff and use any open-source model.
- Model guide
Search, filter and compare open coding models to use with CoderBaba by size, architecture, context window and license.
- Blog
Tutorials, guides and updates about AI coding agents, tools and open models.
- Conductor alternative
A migration guide for Conductor users: back up before November 20, 2026 and switch to parallel agents with CoderBaba.
- RSS feed
Subscribe in any feed reader to get every new post and model addition.
Open-source coding models (9)
- Qwen3-CoderAlibaba Qwen
Alibaba's agentic coding model family. A 480B mixture-of-experts flagship with 35B active parameters, 256K native context and an Apache 2.0 license.
- DevstralMistral AI & All Hands AI
A 24B agentic coding model built for software-engineering agents like OpenHands. Apache 2.0, 128K context, and small enough to run on a single RTX 4090 or a 32 GB Mac.
- Qwen2.5-CoderAlibaba Qwen
A dense code model family from 0.5B to 32B parameters. The 32B Instruct model reached GPT-4o-level code generation at release and is easy to run locally.
- DeepSeek-Coder-V2DeepSeek
A mixture-of-experts code model with 236B total and 21B active parameters, 128K context and support for 338 programming languages. A Lite 16B variant is also available.
- CodestralMistral AI
Mistral AI's 22B open-weight code model, trained on 80+ programming languages, with strong fill-in-the-middle support for IDE autocomplete. Non-production license.
- Granite CodeIBM
IBM's enterprise-focused code models from 3B to 34B, trained on 116 programming languages with license-permissible data and released under Apache 2.0.
- CodeGemmaGoogle
Google's lightweight Gemma-based code models (2B and 7B) for fast code completion and generation on laptops and edge devices.
- StarCoder2BigCode (Hugging Face, ServiceNow, NVIDIA)
Fully open code models (3B, 7B, 15B) trained on The Stack v2 with transparent data, published training details and opt-out support for code authors.
- Code LlamaMeta
Meta's Llama 2-based code models from 7B to 70B, with Python-specialized and instruction-tuned variants. One of the first widely adopted open code LLMs.
Blog posts (4)
- Meta Acquires Conductor: What It Means for Developers and What to Do Before Nov 20Nov 5, 2026
Meta has acquired AI coding startup Conductor, and its cloud service shuts down November 20. What happened, what changes and how to protect your work.
- Welcome to CoderBaba: Build Anything from the Command LineNov 5, 2026
CoderBaba is a free, open-source AI coding agent for your terminal that runs any open-source model, locally or in our free cloud. Here is what it does.
- How to Run an Open-Source Coding Model Locally with OllamaNov 4, 2026
A step-by-step guide to running Qwen2.5-Coder, Devstral or DeepSeek-Coder-V2 on your own machine with Ollama and connecting it to your editor.
- How to Choose an Open-Source Coding Model: License, Size and Context ExplainedNov 3, 2026
Parameters, MoE, context windows and licenses all matter when you choose an open coding model. A practical checklist for developers and teams.
Topics (5)
- AI Coding News: Open Models, Tools and Industry Moves2 posts
The latest news on open-source coding models, AI coding tools and agents, including releases, acquisitions and shutdowns, explained for developers.
- AI Coding Tools: Editors, Agents and Assistants Compared2 posts
News, guides and comparisons of AI coding tools, from autocomplete and IDE assistants to terminal agents and parallel-agent orchestrators, with open-source picks.
- Local AI for Coding: Run Models on Your Own Hardware2 posts
Everything about running AI coding models locally, including hardware, runtimes like Ollama, llama.cpp and vLLM, quantization, privacy and performance tips.
- Tutorials: Run and Use Open-Source Coding Models1 post
Step-by-step tutorials for running open-source coding models locally, connecting them to your editor and setting up AI coding agents you control.
- Guides: Choosing and Understanding Open Coding Models1 post
Decision guides and explainers for developers on choosing open-source coding models, licenses, quantization, benchmarks, MoE and local vs. hosted AI.
How CoderBaba is organized
CoderBaba is a free, open-source AI coding agent for the command line. The how it works page covers the tool itself. The rest of the site is built around three kinds of pages, each with a different job.
Model pages are reference material. Each one covers a single open-source coding model or model family: who built it, its sizes and architecture, its context window and license, its reported benchmarks, the hardware it needs, how to run it locally and how it compares with alternatives. Model pages are updated when new versions ship, and each one shows its "last updated" date.
Blog posts are dated. News posts explain releases, tool launches and industry events such as acquisitions and shutdowns. Tutorials walk you through a specific setup step by step. Guides help you make decisions, such as which model to choose or how to read a license.
Topic hubs connect the two. Each hashtag topic gathers every related post and opens with a short primer on the subject, so you can get oriented before diving into individual articles.
Where to start
- New to open coding models? Read the overview on the home page, then the guide to choosing a model.
- Want to run one today? Follow the Ollama tutorial. It takes about five minutes on most laptops.
- Comparing models for a project? Use the filters and comparison table in the directory.
- Leaving Conductor? Start with the Conductor alternative guide and its backup checklist.
- Ready to build? The how it works page covers installing CoderBaba with
npm install -g coderbabaand running your first task.
Glossary of terms used on this site
- Open-weight model: a model whose trained parameters are published for download, so you can run it yourself.
- Parameters (B): the number of learned weights in billions. A rough proxy for capability and memory needs.
- Dense model: uses all parameters for every token it generates.
- Mixture of experts (MoE): activates only some "expert" sub-networks per token. Total size sets memory needs, and active size sets speed.
- Context window: how many tokens the model can consider in one request, including your prompt, files and its reply.
- Quantization: storing weights at lower precision, such as 4-bit, to reduce memory with a small quality cost.
- Fill-in-the-middle (FIM): a training mode that lets a model complete code using text both before and after the cursor, which is essential for autocomplete.
- Agent: an AI system that takes actions such as reading files, editing code and running commands in a loop until a task is done.
- Git worktree: an extra checkout of a repository on its own branch, used to give parallel agents isolated workspaces.
- HumanEval: a benchmark of 164 Python function-writing problems checked by unit tests.
- SWE-bench Verified: a benchmark of real GitHub issues where a model must produce a patch that passes hidden tests.
- OpenAI-compatible API: a widely used HTTP interface for chat models. Most local runtimes offer it, so tools can switch between hosted and local models.
Can't find something?
Every page on CoderBaba appears in the lists above. If you're looking for a specific model, use the search and filters in the model directory. If a model or topic you expected is missing, it probably hasn't been covered yet. New models are added as they're released and tested.
For search engines and feed readers
A machine-readable version of this index is available at /sitemap.xml and updates automatically whenever a model or post is added. New posts are also published to the RSS feed.