HalCode AI Assistant

Your conversational AI coding partner for VS Code.

Plan once → approve once → build full projects in an indexed folder with real tools. Works with Claude, Gemini, Grok, and Ollama (local PC or server, plus Ollama Cloud).

We're back (v4.9.0)

HalCode was quiet for months. We’re back with a much stronger core: not three competing modes and fragile markdown scraping, but one agent that feels like working with someone — plans, builds, fixes missing pieces in chat, and helps you run the stack.

More updates are coming — especially around newer cloud models and local / server Ollama models.

How it works now

1. Point HalCode at a folder

Use Configure Indexed Folders. HalCode may only create/read/delete under those paths.

2. Chat like a teammate

Examples:

Create a fullstack pizza site with TypeScript, React, Vite, and a Node API
Fix the backend auth routes and add missing files
Read d:\myapp\README.md and summarize setup

3. Plan → one approval

HalCode outlines the approach and lists files. You Approve (task panel or a clear yes).

4. It builds with tools

It uses structured tools (not “guess files from the essay”):

Tool	Purpose
`write_file`	Create/update files
`read_file`	Read project files
`list_dir`	List folders
`run_command`	Install / test / dev (you approve each command)
`delete_file`	Delete (you must confirm in a dialog)

You’ll see progress in chat as files are written.

5. After the build

Optional buttons:

Install dependencies
DB migrate / seed (when Prisma is detected)
Start dev servers
Skip

Every command still asks for confirmation.

6. Keep going

Fix bugs, add features, ask it to read or list files — same chat, same agent.

Buttons that matter

Button	What it does
Send	Your message (agent + history + tools)
Send + Context	Same, plus the active editor file
Rules	Apply your coding rule set
Clear	Clear the chat view

~~Agentic / Workflow / Multi-Agent~~ were removed — they duplicated the main agent and confused the product.

What's new in 4.9.0

Tool-first agent for Claude, Gemini, Grok, Ollama local & cloud
Full-app builds (dozens of files) with live write progress
Post-build run helpers (install / migrate / dev)
Delete with approval
Model catalogs updated (Grok 4.x family, modern Ollama tags)
Simpler UI — one agent path
Reliability: no hanging “thinking” on normal Q&A; history trim; safer paths

See CHANGELOG.md for the full list.

Quick start

Install the extension and reload VS Code.
Set an API key (or Ollama) in HalCode settings.
Configure Indexed Folders → pick your project directory.
Open HalCode chat.
Describe a project or a change → approve → watch it build.

Providers

Claude (Anthropic)
Gemini (Google)
Grok (xAI)
Ollama local (localhost:11434) or your server
Ollama Cloud

Prefer coding-capable models for multi-file builds (e.g. strong Grok / Claude / Gemini, or qwen2.5-coder locally).

Local Ollama tip

ollama pull qwen2.5-coder:7b

Then select that model in settings and set Ollama location to local (or server).

Design principles (4.9+)

One agent — chat is the product.
Tools over regex — file ops are deterministic; fences are fallback only.
Indexed sandbox — writes/deletes stay inside folders you chose.
You stay in control — approve plans, commands, and deletes.
Conversational follow-through — missing files and fixes stay in the same thread.

License & trial

See in-extension licensing for Free / Pro trial details. The goal is a useful free path and a fair Pro path as we grow.

Feedback & roadmap

Issues and ideas welcome on the project repo.
Near-term: more model support, local-model polish, packaging, and long-build reliability.

Thanks for sticking with HalCode — or for finding us again.
We’re building in the open, one solid agent loop at a time.

Older releases

History from 4.8.x and earlier remains in CHANGELOG.md (parsing fixes, codeblocks, MCP, embeddings, etc.).

Genius Coding Assistant

Hal