FZ-Craft Code
An agentic AI coding assistant for VS Code. It is connected to FZ-Craft by default and works with any OpenAI-compatible API: OpenAI, OpenRouter, Google Gemini, DeepSeek, Groq, Together, a local Ollama or LM Studio, vLLM, llama.cpp and others. You bring your own API key.
Quick start (FZ-Craft)
- Install the extension. FZ-Craft is already selected (
https://fzcraft.pikonox.com/v1, default model deepseek-v4.1-flash).
- When the welcome prompt appears, click Set API Key and paste your FZ-Craft key (it starts with
fz_live_). Get a Key signs you in and opens Dashboard → API keys. The key is checked as soon as you save it.
- Open the chat with the FZ icon in the activity bar, or press
Cmd+Alt+O / Ctrl+Alt+O.
- Click the model chip to pick a model. The list comes live from FZ-Craft's catalog (
/models.json), so it works even before you add a key:
- Free models (
deepseek-v4.1-flash, glm-5.3) are at the top.
- Models that need a paid plan show a lock and how to unlock them (Starter plan or Day Pass). Unlock more models opens the billing page.
- Once your key is set, the list follows your key's plan (from
/v1/models).
The composer bar
┌───────────────────────────────────────────────────────────────┐
│ Ask anything, / for commands, + to add context │
│ [+] [/] [🕐 54m] [DeepSeek V4.1 Flash High] [⚡ Auto] [↑] │
└───────────────────────────────────────────────────────────────┘
| Button |
What it does |
| + |
Attach context: the current file, the editor selection, workspace files, or the problems in the current file |
| / |
Command menu. You can also type / in the message box |
| 🕐 54m |
How long the current session has been running. Click it to browse, resume or delete saved chats |
| Model chip |
Shows the model's name. Click it to choose a model from the provider's live list (search by name or ID, or type any model ID), set the reasoning effort (Auto, Low, Medium, High, Extra high), switch provider, or set the API key |
| ⚡ Mode |
Permission mode (see below). Shift+Tab in the message box cycles through the modes |
| ↑ / ■ |
Send the message (Enter). While the agent is working, the button stops it (Esc) |
Permission modes
| Mode |
Behaviour |
| Ask (default) |
Every file change opens a diff and waits for Apply or Reject. Commands wait for Run |
| Auto-edit |
File changes are applied automatically; commands still ask first |
| ⚡ Auto |
Files are edited and commands run without asking |
| Plan |
Read-only: the agent explores the code and proposes a step-by-step plan without changing anything |
Slash commands
| Command |
Description |
/new |
Start a new chat |
/history |
Browse and resume past chats |
/model |
Choose the model and reasoning effort |
/mode |
Change the permission mode |
/provider |
Switch AI provider |
/key |
Set the API key |
/compact |
Summarize the chat to free up context (useful in long sessions) |
/init |
Analyze the project and write an AGENTS.md |
/review |
Review uncommitted changes (git diff) |
/explain |
Explain the selection or current file |
/fix |
Fix errors and warnings in the current file |
/test |
Write and run tests for the selection or current file |
/settings, /help |
Open settings, or show tips |
Text after a command is passed along as extra instructions, for example /fix the null check in login().
Features
- Agentic loop: the model explores your project, reads and edits files, runs commands and checks diagnostics until the task is done.
- Diff approvals in VS Code's diff editor, with an Always allow option for the rest of a chat.
- Saved chat history per workspace. Resume any past chat from the 🕐 menu.
- Reasoning effort: sent as
reasoning_effort. If a model rejects it, the request is retried without it automatically.
- Clear errors: API error codes are turned into next steps, for example this model needs a paid plan, pick a Free model, token allowance used up, key spend limit reached or key paused.
- Automatic retries: rate limits (
429, waiting for retry-after) and temporary upstream failures (500/502) are retried with a notice in the chat. Quota and plan errors are not retried.
- Capability-aware: if the catalog says a model can't call tools, the agent answers without them instead of failing.
- Secure keys: API keys are stored in VS Code's SecretStorage (the OS keychain), never in
settings.json.
- Streaming, including the "thinking" output of reasoning models.
- Editor-aware: the model knows the open file and selection.
Cmd+Alt+J / Ctrl+Alt+J attaches the selection.
- Project instructions:
AGENTS.md in the workspace root is added to the system prompt.
| Tool |
What it does |
Needs approval (Ask mode) |
read_file, list_directory, find_files, search_text, get_diagnostics |
Explore the code |
no |
edit_file, write_file |
Change or create files |
yes (diff shown) |
run_command |
Run a shell command in the workspace root |
yes |
All file access is restricted to the open workspace folder. Plan mode only offers the read-only tools.
Providers
FZ-Craft comes first in the list and is selected on install. To add another provider, use FZ-Craft Code: Add Provider or edit settings.json:
"fzCraft.providers": [
{
"name": "FZ-Craft",
"baseUrl": "https://fzcraft.pikonox.com/v1",
"model": "deepseek-v4.1-flash",
"apiKeyUrl": "https://fzcraft.pikonox.com/login?next=%2fdashboard%2fapi-keys"
},
{ "name": "Ollama (local)", "baseUrl": "http://localhost:11434/v1", "model": "qwen2.5-coder:7b" },
{
"name": "My vLLM",
"baseUrl": "http://gpu-box:8000/v1",
"model": "my-model",
"supportsTools": false, // chat-only models
"maxTokens": 4096,
"headers": { "X-Custom": "abc" } // extra HTTP headers
}
],
"fzCraft.activeProvider": "FZ-Craft"
baseUrl is the URL that /chat/completions and /models are appended to. API keys are stored per provider name, so if you rename a provider, set its key again.
Settings
| Setting |
Default |
Description |
fzCraft.defaultMode |
ask |
Starting permission mode. Your ⚡ choice is remembered after that |
fzCraft.showDiffOnApproval |
true |
Open a diff automatically for proposed edits |
fzCraft.maxIterations |
25 |
Maximum tool-use steps per message |
fzCraft.temperature |
provider default |
Sampling temperature |
fzCraft.includeEditorContext |
true |
Tell the model about the open file and selection |
fzCraft.customInstructions |
"" |
Extra instructions added to the system prompt |
Requirements
- VS Code 1.90 or newer.
- An API key for your provider. FZ-Craft has a free tier with free models (get a key). Local servers like Ollama and LM Studio need no key.
- A trusted workspace folder. The agent edits files and runs commands, so the extension is disabled in Restricted Mode.
Privacy and security
- Your messages, and any file contents or command output the agent reads, are sent only to the provider you choose (FZ-Craft by default) to generate answers. Nothing is sent anywhere else, and the extension has no telemetry.
- API keys are stored in VS Code's SecretStorage (your OS keychain), never in
settings.json.
- The agent's file tools can only read and write inside the open workspace folder. Shell commands run in the workspace folder with your user's permissions. In Ask mode (the default), every file change and every command waits for your approval.
- Chat history is saved locally in VS Code's storage for the extension.
Support
Visit fzcraft.pikonox.com for plans and models, and the API docs for details on the FZ-Craft API.
License
Proprietary. Copyright (c) 2026 Pikonox. All rights reserved. The full terms of use are in the LICENSE file included with the extension.
| |