SideCar — AI Coding Assistant for VS Code

Fully Offline· Autonomous Agent Loop· 86 Built-in Tools· MCP Server Support· Security Scanning· Dep Drift Alerts· Model Arena· SIDECAR.md Retrieval· Code Profiling· Inline Diff Preview· Git Integration· Smart Context· OpenAI-Compatible· Cycle Detection· .sidecar Directory· MIT License· Zero Subscriptions· Multi-file Edit Streams· Regression Guards· RAM/VRAM Monitor· Speculative FIM· Skills Registry· Skills Picker UI· Fully Offline· Autonomous Agent Loop· 86 Built-in Tools· MCP Server Support· Security Scanning· Dep Drift Alerts· Model Arena· SIDECAR.md Retrieval· Code Profiling· Inline Diff Preview· Git Integration· Smart Context· OpenAI-Compatible· Cycle Detection· .sidecar Directory· MIT License· Zero Subscriptions· Multi-file Edit Streams· Regression Guards· RAM/VRAM Monitor· Speculative FIM· Skills Registry· Skills Picker UI·

Not a
chatbot
plugin.

SideCar runs an autonomous loop — closer to Claude Code or Cursor than anything else in the VS Code ecosystem. It decides what to read, what to change, runs tests, and iterates. Use local Ollama models, the Anthropic API, or any OpenAI-compatible server.

Core capability

Autonomous Agent Loop

Multi-step reasoning with real tool use. SideCar reads the files that matter, writes targeted diffs, runs your test suite, reads the output, and fixes what broke — all in one uninterrupted pass. Cycle detection prevents runaway loops.

file read/write test runner iterative repair diff preview cycle detection database query visual verification doc-to-test dep drift alerts model arena

Smart Context

AST-based extraction for JS/TS, Python, Rust, Go, and Java/Kotlin. Sends relevant functions, not entire files.

Security Scanning

Built-in secrets detection and vulnerability patterns. Scans staged files before every commit.

Dependency Drift Alerts

Scans package.json, requirements.txt, Cargo.toml, and go.mod for outdated and vulnerable deps. CVE/GHSA IDs from OSV surface in the Problems panel.

Model Arena

Compare 2–4 models side-by-side on the same prompt. Local ELO ratings, multi-turn chat, and agent-mode (fork dispatch per model).

MCP Servers

Extend with Model Context Protocol servers — databases, APIs, browser automation, and more.

Git Integration

Stage, commit, push, branch, and stash through natural language. Commit messages auto-generated.

Multi-file Edit Streams

For wide refactors, SideCar plans a DAG of edits before touching anything — shows every file, op, and rationale upfront. Files stream in parallel; cancel any file mid-write without disturbing the rest.

Regression Guards

Shell-command guards wired into the agent loop. Built-in guards (lint-clean, tests-pass, no-new-todos) run after edits and block the turn on failure. Enable per-skill via guards: frontmatter.

Change-Impact Analysis

A consequence-aware code graph (AST-exact call/type-flow edges for TS + Python). analyze_impact answers "what depends on this symbol?" — import-resolved callers, type-users, subtypes, importers — and an opt-in gate blocks finishing with unverified cross-file dependents. For scientific Python, check_numerical_contracts + check_shape_consistency flag array kernels missing a shape/dtype contract and propagate shape specs to catch provable conflicts.

The only free, local
VS Code extension
with a full
agent loop.

Tested against Continue,
Llama Coder, Twinny,
Copilot Free & Pro,
and Claude Code.

Capability	SideCar	Continue	Llama Coder	Twinny	Copilot Free	Copilot Pro	Claude Code
Chat with local models	✓	✓	—	✓	✓	—	—
Inline completions	✓	✓	✓	✓	✓	✓	—
Autonomous agent loop	✓	—	—	—	—	✓	✓
File read / write / edit	✓	—	—	—	—	✓	✓
Run commands & tests	✓	—	—	—	—	✓	✓
Security & secrets scanning	✓	—	—	—	—	—	—
MCP server support	✓	—	—	—	—	✓	✓
Hooks & scheduled tasks	✓	—	—	—	—	—	✓
Git integration	✓	—	—	—	—	✓	✓
Diff preview & rollback	✓	—	—	—	—	✓	✓
Diagram generation	✓	—	—	—	—	✓	—
OpenAI-compatible API	✓	✓	—	—	—	—	—
Fully offline	✓	✓	✓	✓	—	—	—
Free & open source	✓	✓	✓	✓	Freemium	$10+/mo	API costs

Four steps
to agentic coding.

Install Ollama

Download from ollama.com — the local model runtime SideCar uses by default. Make sure it's in your PATH.

Install SideCar

Search sidecar-ai in the VS Code Extensions panel, or install directly from the Marketplace.

Open the Panel

Click the SideCar icon in the activity bar. The extension launches Ollama automatically on first run.

Give it a task

Describe what you want done. SideCar will read the relevant files, write the code, and test it — no hand-holding required.

Requirements

Ollama

Installed and in PATH. Runs your local language models.

VS Code 1.88.0+

Any current release. VS Code Insiders also works.

Anthropic API (optional)

Use Claude models via API key. Prompt caching for lower cost.

OpenAI-compatible backend (optional)

LM Studio, vLLM, Groq, OpenRouter — point SideCar at any compatible API.

Your code.Your machine.No cloud.

Not achatbotplugin.