# GBS SE Agent

> An open-source coding agent for VS Code. Describe a task and it explores your codebase, edits files and runs commands, showing every step. Runs on local models, so your code can stay on your machine.

GBS SE Agent · Start · Source: https://kit.gramproindia.com/se-agent · Code: https://github.com/gramproindia/agent

GBS SE Agent is a coding agent that lives in the VS Code sidebar. You describe
the work, and it does it the way a careful colleague would: it searches and
reads before it changes anything, asks before anything risky, and shows you
every file it touched when it finishes.

Point it at a model running on your own machine (Ollama, LM Studio,
llama.cpp or vLLM) and no code leaves the laptop. Cloud endpoints work too,
and the agent asks before it sends anything to one.

![Six steps in a row: you describe the job in your own words; it looks, searching and reading the files; it asks before anything risky; it changes the code; it checks, running the tests and reading errors; and it shows you every file it changed. An arrow runs from the checking step back to the looking step: if a check fails, it goes back and fixes it, as many times as it takes.](https://kit.gramproindia.com/diagrams/se-agent-plain-light.webp)

_One instruction becomes a loop of small steps. The loop back from checking is what makes it an agent rather than a chatbot._

## What it does well

- **Runs on local models.** Ollama is the built-in default. Any server with an
  OpenAI-compatible API works, local or cloud.
- **Shows its work.** Every search, read, edit and command appears as it
  happens, with live command output and a diff for every change.
- **Asks first.** Shell commands, and edits to files that run code or
  configure tooling, wait for your approval. A repository cannot change where
  your code is sent or switch the approvals off.
- **Learns your stack.** [Skills](https://kit.gramproindia.com/se-agent/skills) teach it your component
  library and project rules, loaded only when a task needs them.
  [MCP servers](https://kit.gramproindia.com/se-agent/mcp-servers) give it tools, such as the GramproKit
  catalogue.
- **Lets you take it back.** Each turn ends with a summary of every changed
  file and an **Undo this turn** button.

## Start here

| If you want to | Read |
| --- | --- |
| Install it and run a first task | [Getting started](https://kit.gramproindia.com/se-agent/getting-started) |
| Keep code on your machine | [Local models](https://kit.gramproindia.com/se-agent/local-models) |
| Use a cloud model or a free tier | [Cloud models](https://kit.gramproindia.com/se-agent/cloud-models) |
| Get the most out of a session | [Working with the agent](https://kit.gramproindia.com/se-agent/using-the-agent) |
| Build UI with GramproKit components | [Using it with GramproKit](https://kit.gramproindia.com/se-agent/with-gramprokit) |
| Understand how it works inside | [Architecture](https://kit.gramproindia.com/se-agent/architecture) |
| Explain it to someone non-technical | [In plain English](https://kit.gramproindia.com/se-agent/plain-english) |

## At a glance

| | |
| --- | --- |
| Runs in | VS Code 1.120 or later, in trusted workspaces |
| Models | Ollama, LM Studio, llama.cpp, vLLM, Hugging Face, Groq, Google AI Studio, any OpenAI-compatible endpoint |
| Built-in tools | read, edit and write files; search; list folders; run commands; a task checklist; subagents; skills |
| Extensions | Skills (`SKILL.md` folders) and MCP servers |
| Source | [gramproindia/agent](https://github.com/gramproindia/agent), MIT licensed, no runtime dependencies |