A local-first coding agent powered by Ollama LLMs
Privacy-first AI that runs entirely on your machine. An agentic loop that reads, searches, edits, and runs your build and tests to verify its own work, plus on-demand skills, tool calling, and spec-first scaffolding.
$ devbldr /path/to/project
Indexing project... done
Found 42 files, 156 symbols
> Add authentication to the user API
⏺ read user routes, edited handler, ran tests ✓_
devBldr uses Ollama to run AI models locally, keeping your code private while an agentic loop plans, writes, refactors, and verifies your code with the safety gates always in your hands.
A single loop that reads, searches, edits, and runs your build and tests to verify its own work. Follow-ups keep their context.
Instead of stuffing the whole project into every prompt, the model requests what it needs (read, search, edit, and run commands) on demand.
On-demand SKILL.md instruction folders, loaded only when a task matches, so specialized know-how stays out of the prompt until it is relevant.
Point it at an empty directory and it plans a complete, runnable app (requirements, architecture, per-file generation), then builds and verifies it, all with your approval.
Uses Ollama for privacy-first AI inference. Your code never leaves your machine.
Automatically indexes your codebase using tree-sitter and builds intelligent context for each request.
Every write goes through colorized diffs, an approval workflow, and automatic backups. Commands ask before they run. The agent works with the gates, never around them.
Tree-sitter parsing for C, C++, Python, JavaScript, TypeScript, JSX, TSX, Rust, Go, Java, and more.
Recognizes Flask, Django, FastAPI, Express, NextJS, React, Vue, C++, Rust, and Go projects.
On startup, devBldr indexes your project with tree-sitter to extract symbols and detect the project type. What happens next depends on the request.
Fixing a bug, adding a feature, refactoring: devBldr hands the model the full tool set and lets it drive. It understands, changes, and runs your build and tests to verify, looping until the task is done.
After generating code, devBldr runs your build and test suite, reads any failures, and keeps fixing and re-running until everything passes.
Your request expands into requirements, then an architecture, then a file plan you approve. Each file is generated, then the result is built and exercised automatically.
Get started in seconds on your Mac
# Add the tap and install
$ brew tap jefferyabbott/devbldr
$ brew install devbldr
# Or install in one command
$ brew install jefferyabbott/devbldr/devbldr
Make sure you have Ollama installed and running
Just type what you want in plain language and the agent handles the rest. These slash commands are there when you want them.
Apple Silicon Mac
macOS 12.0 or later
Ollama installed and running
Agentic features work best with a tool-capable model (llama3.1/3.2/3.3, qwen2.5/qwen3, mistral-nemo).
If you find devBldr useful, consider buying the developer a coffee!
☕ Buy Me a Coffee