User Manual & Reference

How to use Limitcode Pro.

Everything you need to install Limitcode in Sublime Text, connect your preferred AI models, collaborate on real codebases, review diffs safely, and customize the agent for your workflow.

Getting Started

What Limitcode Does

Limitcode Pro is an autonomous AI coding assistant that lives directly inside Sublime Text 4 (builds 4050+). Instead of switching to a heavy Electron IDE or copying snippets into a browser, you open a native Sublime side-by-side view where the assistant can work directly with your project.

From a single conversation inside Sublime Text, you can ask Limitcode to:

  • Explore and explain your codebase: list folders, search for symbols or text across files, and read implementation details.
  • Write and refactor code: create new files, make targeted edits to existing functions, or apply multi-file patches.
  • Review every change safely: inspect proposed changes in Sublime's native diff view and undo or redo any edit in one keystroke (Ctrl+Alt+U).
  • Run commands and tests: execute your build scripts, test suites, linters, or git commands — and keep long-running dev servers running in the background.
  • Look up docs & inspect screenshots: search the live web, fetch documentation pages, or analyze screenshots pasted directly into the chat (Ctrl+V).
  • Follow your team's workflow: automatically load repository rules (AGENTS.md), reusable Agent Skills, external MCP servers, and specialized Subagents.
Getting Started

Installation & License Activation

Limitcode Pro is distributed as a self-contained .sublime-package file. No external daemon or Python environment setup is required.

Step 1

Download your package

After purchasing on Gumroad, copy your license key from the receipt and download the Limitcode.sublime-package file matched to your Sublime Text 4 build.

Step 2

Place in Installed Packages

In Sublime Text, click Preferences → Browse Packages… (or Sublime Text → Settings → Browse Packages… on macOS). Go one folder up to open the Installed Packages directory, drop Limitcode.sublime-package inside, and restart Sublime Text.

Step 3

Activate your license

Open the Command Palette (Ctrl+Shift+P on Windows/Linux or Cmd+Shift+P on macOS), run Limitcode Pro: Activate License, and paste your license key.

Managing your 3 devices & working offline: Your license can be active on up to 3 personal computers at the same time. You can check your seat status anytime with Limitcode Pro: Show License Status, or free up a seat to move to another computer using Limitcode Pro: Deactivate This Device. Once activated, you can work offline for up to 30 days between periodic checks.
Getting Started

Connecting an AI Provider

Limitcode supports over 40 AI providers. You can connect models in four different ways depending on what you already use:

1. Direct API Key (Fastest)

Press Ctrl+Alt+K or run Limitcode: Setup Provider API Key, choose your provider — such as OpenRouter, Anthropic, OpenAI, Google Gemini, DeepSeek, Mistral, Groq, or xAI — and paste your API key.

2. Account Sign-In (OAuth)

Use an existing subscription without creating an API key. Run Limitcode: Login with GitHub Copilot with Ctrl+Alt+B, Login with Codex with Ctrl+Alt+Z, or Login with Google Antigravity with Ctrl+Alt+Shift+G and complete the browser sign-in.

3. Existing CLI Login

If you already use Claude Code CLI or Codex CLI in your terminal and are logged in, Limitcode can reuse that session directly. Run Limitcode: Login with Claude Code with Ctrl+Alt+E or select the CLI provider.

4. Local Models (Ollama & LM Studio)

To run models 100% locally on your machine, start Ollama at http://localhost:11434/v1, LM Studio at http://localhost:1234/v1, or LiteLLM, then select it with Ctrl+Alt+P via Limitcode: Change Provider.

Complete List of Supported Providers

Below is the full catalogue of AI providers supported in Limitcode, along with each provider's Settings ID for Limitcode.sublime-settings and supported models:

Recommended

Quick Setup via Command Palette

Press Ctrl+Alt+K (or Cmd+Alt+K on macOS) to run Limitcode: Setup Provider API Key, select your provider from the list, and paste your API key. Limitcode stores it securely in your settings.

Manual JSON

Configure in Sublime Settings

Open Limitcode.sublime-settings and add the provider's Settings ID under the "api_keys" object: "api_keys": { "<settings_id>": "your-key" }.

Direct API Keys — Frontier & Cloud Providers

All direct providers connect via your personal API key. Set with Ctrl+Alt+K or define inside "api_keys" in Limitcode.sublime-settings.

Provider Settings ID Supported Models & Configuration
OpenRouter Gateway
openrouter
Unified access to 500+ models across Anthropic, OpenAI, Meta, DeepSeek, and more with a single account.
Endpoint: openrouter.ai/api/v1
Anthropic Frontier
anthropic
Direct official access to Claude Sonnet 5, Opus 5.5, and Fable 5.1 with prompt caching.
OpenAI Frontier
openai
Direct official access to GPT-6, GPT-6 Astra, GPT-6 Sol, and o-series reasoning models.
Google Gemini Direct API
gemini
Google AI Studio API key with direct access to Gemini 4 Argon and Gemini 3.7 Flash.
DeepSeek Direct API
deepseek
DeepSeek V4, R1 reasoning series, and specialized coder models.
Endpoint: api.deepseek.com
xAI (Grok) Direct API
xai
Grok 4.20 and high-context coding/reasoning models.
Endpoint: api.x.ai/v1
Mistral AI Direct API
mistral
Mistral Large, Codestral, Devstral, and open-weight models.
Endpoint: api.mistral.ai/v1
Groq Fast LPU
groq
Ultra-fast inference powered by LPU hardware for open-weight models.
Endpoint: api.groq.com/openai/v1
Cerebras Fast Wafer
cerebras
High-speed wafer-scale engine for ultra low-latency code completion.
Endpoint: api.cerebras.ai/v1
Together AI Cloud Host
together
Cloud-hosted open-weight chat and code completion models.
Endpoint: api.together.xyz/v1
DeepInfra Serverless
deepinfra
Serverless pay-per-token inference for open-source AI models.
Endpoint: api.deepinfra.com/v1/openai
Moonshot AI / Kimi Direct API
moonshot
Kimi K2.5 and ultra-long context reasoning models.
Endpoint: api.moonshot.ai/v1
Alibaba DashScope (Qwen) Direct API
alibaba
Qwen3 Coder and complete Qwen open model family.
Endpoint: dashscope.aliyuncs.com/compatible-mode/v1
Xiaomi MiMo Direct API
xiaomi
Xiaomi MiMo reasoning and code synthesis models.
NVIDIA NIM Cloud Host
nvidia
Optimized enterprise inference containers hosted on NVIDIA infrastructure.
Endpoint: integrate.api.nvidia.com/v1
Perplexity Search AI
perplexity
Sonar live search-augmented intelligence and reasoning models.
Endpoint: api.perplexity.ai
Venice AI Privacy
venice
Zero-data-retention, privacy-first private inference.
Endpoint: api.venice.ai/api/v1
Cohere Enterprise
cohere
Command R+ and enterprise language and embedding models.
OpenCode Zen Coding Hub
opencode-zen
Curated gateway optimized specifically for developer coding workflows.
Endpoint: opencode.ai/zen/v1
OpenCode Go Low Latency
opencode-go
Ultra-fast coding gateway (can also share your OpenCode Zen key).
Endpoint: opencode.ai/zen/go/v1

Enterprise Cloud, Gateways & Custom Endpoints

Connect corporate cloud deployments, AI observability proxies, or any custom OpenAI-compatible server.

Provider Settings ID Supported Models & Configuration
Amazon Bedrock AWS Cloud
bedrock
AWS Bedrock Converse API for VPC-hosted Claude, Llama, and Titan models inside your AWS perimeter.
Azure OpenAI Azure Cloud
azure
Private GPT and reasoning deployments on Microsoft Azure Cognitive Services.
Google Cloud Vertex AI GCP Cloud
vertex
Enterprise Google Cloud Vertex AI native endpoint for Gemini models.
Vertex AI Anthropic GCP Cloud
vertex_anthropic
Claude models hosted natively within Google Cloud Vertex AI infrastructure.
Cloudflare Workers AI Edge
cloudflare
Serverless edge inference distributed across Cloudflare's global data centers.
Endpoint: api.cloudflare.com/client/v4/accounts/{id}/ai/run/{model}
Vercel AI Gateway Gateway
vercel
Unified proxy, rate-limiting, and telemetry for frontend deployments.
Endpoint: ai-gateway.vercel.sh/v1
Helicone Observability
helicone
LLM observability, analytics, caching, and rate-limiting proxy.
Endpoint: oai.hconeai.com/v1
GitLab AI Enterprise
gitlab
GitLab Duo enterprise AI token for repository and code tasks.
Endpoint: gitlab.com/api/v4/ai
SAP AI Core Enterprise
sap
Enterprise AI credentials for SAP business cloud environments.
Endpoint: api.ai.sap.com/v1
Custom OpenAI-Compatible Self-Hosted / Proxy
openai-compatible / custom
Connect any self-hosted proxy, vLLM, TGI, or custom enterprise gateway. Set custom URL in provider_base_urls.

Account Sign-In (OAuth) & Local CLI Sessions

Use existing subscriptions or authenticated terminal sessions without creating separate API billing keys.

Provider Settings ID Supported Models & Configuration
GitHub Copilot OAuth Login
copilot
Authorize via browser with your GitHub subscription. Run Limitcode: Login with GitHub Copilot via Ctrl+Alt+B.
OpenAI Codex OAuth OAuth Login
codex-oauth
Sign in with your OpenAI account in the browser. Run Limitcode: Login with Codex via Ctrl+Alt+Z.
Google Antigravity OAuth OAuth Login
google-antigravity
Sign in with your Google account via browser. Run Limitcode: Login with Google Antigravity via Ctrl+Alt+Shift+G.
Claude Code CLI CLI Session
claude_code_cli
Reuses your active terminal session from 'claude login'. Connect via Ctrl+Alt+E (Limitcode: Login with Claude Code).
Codex CLI CLI Session
codex_cli
Reuses an active terminal session from 'codex login'. Automatically syncs authentication.

Local & Self-Hosted Runtimes (Offline-Capable)

Run open-weight models locally on your GPU/CPU with 100% privacy and zero API token costs.

Provider Settings ID Supported Models & Configuration
Ollama 100% Local / Free
ollama
Connects to your local Ollama instance. Fully offline and private.
Endpoint: http://localhost:11434/v1
LM Studio 100% Local / Free
lm-studio
Connects to LM Studio's local server runtime for GUI-managed local weights.
Endpoint: http://localhost:1234/v1
LiteLLM Local Proxy
litellm
Connects to a local LiteLLM load balancer or proxy server.
Endpoint: http://localhost:4000/v1

Configuring Multiple Providers & Custom Base URLs

You can configure as many providers at once as you like inside Preferences → Package Settings → Limitcode → Settings. When you open the unified model selector with Ctrl+Alt+M, Limitcode combines the available models from all your configured providers into a single list and automatically updates both default_provider and default_model when you pick one:

{
  "default_provider": "openrouter",
  "default_model": "anthropic/claude-sonnet-5",
  "api_keys": {
    "openrouter": "sk-or-...",
    "anthropic": "sk-ant-...",
    "openai": "sk-...",
    "gemini": "AIza...",
    "deepseek": "sk-...",
    "groq": "gsk_...",
    "xai": "xai-...",
    "openai-compatible": "sk-custom-..."
  },
  "provider_base_urls": {
    "openai-compatible": "https://api.example.com/v1",
    "ollama": "http://localhost:11434/v1"
  }
}

Want to compare billing, privacy, and account boundaries across connection methods? Read the dedicated Providers Guide →

Getting Started

Running Your First Task

  1. Open the chat panel: Press Ctrl+Alt+L or run Limitcode: Open Chat. A native side-by-side chat view opens next to your code.
  2. Pick your model: Press Ctrl+Alt+M or run Limitcode: Change Model to choose from all models available across your configured providers.
  3. Ask a question or request an edit: Type a prompt such as "Explain how authentication works in @src/auth.py and add unit tests for token expiry" and press Enter.
  4. Approve or undo: When Limitcode proposes a file edit or shell command, approve it from the quick prompt. If you want to revert an edit at any point, press Ctrl+Alt+U.
Using the Chat

The Chat Panel & Session Management

Limitcode's chat is a native Sublime Text view designed for keyboard-first navigation. Every conversation is automatically saved per project so you can close Sublime and pick up right where you left off.

Managing conversations

Start a fresh conversation anytime with Ctrl+Alt+N via New Chat. Browse and reopen previous sessions for your project with Ctrl+Alt+H via Chat History, give a session a memorable title with Ctrl+Alt+Y via Rename Session, or remove old chats with Ctrl+Alt+D via Delete Session.

Prompt history & cancelling runs

Inside the chat input box, press the ↑ and ↓ arrow keys to cycle through your previously sent prompts. If a response is taking too long or heading in the wrong direction, press Ctrl+Alt+X to cancel the active request, or press Ctrl+Alt+Shift+X or Escape to stop the agent immediately. The chat stays ready for your next instruction.

Code blocks, live checklists & status bar

Code snippets generated in the chat include quick actions to copy the block, insert it into your active editor, or create a new file. When the agent tackles a multi-step task, it displays a Live Task Checklist in the chat so you can follow its progress step by step. At the bottom of the window, the Sublime status bar always shows your active provider, model, work mode, reasoning effort, context usage, and session cost.

Customizing font size & model thoughts

Want the chat text slightly smaller or larger than your code editor? Run Limitcode: Increase Chat Font Size, Decrease Chat Font Size, or Reset Chat Font Size from the Command Palette. You can also show or hide the model's internal reasoning process anytime by pressing Ctrl+Alt+T or running Toggle Show Thoughts.

Using the Chat

Passing Files & Code Context

You never have to manually copy and paste code between your editor and the chat. Limitcode gives you three ways to point the agent at the exact code you care about:

1. Mention files with @

Type @ directly inside the chat input to reference a file or path in your workspace, such as @src/components/Navbar.tsx. Limitcode attaches the file path so the agent knows exactly which file to inspect or modify.

2. Send Selection or Active File

Highlight a function, class, or block of lines in any open editor tab and press Ctrl+Alt+A or run Limitcode: Send to Agent. If nothing is highlighted, pressing Ctrl+Alt+A sends the entire active file. The code is attached to your chat prompt immediately.

3. Let the agent search on its own

Don't know which file handles a feature? Just ask in plain language — for example, "Where do we validate webhook signatures?" Limitcode will search your repository, list relevant directories, and read the matching files automatically.

Tip — Saved files vs. unsaved buffers: When Limitcode reads or searches files on its own, it reads the saved files on disk. If you have unsaved edits in an open tab that you want the agent to see right now, either save the file first with Ctrl+S or press Ctrl+Alt+A to send the live buffer.
Using the Chat

Pasting Screenshots & Configuring Web Search

1. Inspect images & screenshots

Copy any screenshot, UI glitch, terminal error dialog, database diagram, or Figma mockup to your clipboard and press Ctrl+V or Cmd+V on macOS directly inside the chat. Vision-capable models such as Claude Sonnet 5, Claude Opus 5.5, GPT-6, and Gemini 4 will analyze the image alongside your code.

2. Built-in Web Search (Works out of the box, no setup required)

You can ask Limitcode to search the web for up-to-date documentation, library changelogs, or error messages — or ask it to fetch and read a specific URL. Both actions ask for your approval before making external requests.

Default zero-config search: Even if you haven't configured any search API key, Limitcode includes a built-in search that works out of the box using DuckDuckGo or the Chinese search engine Sogou. While convenient for quick lookups without setup, these free default engines are basic and not always the most accurate for deep technical queries.

3. How to configure Brave Search or Tavily (Recommended for best results)

For much higher quality and more reliable technical search results, configure Brave Search or Tavily inside your Limitcode settings by pressing Ctrl+Alt+S or opening Preferences → Package Settings → Limitcode → Settings, and adding the "web_search" block:

{
  "web_search": {
    "provider": "auto",
    "brave_api_key": "BSA...",
    "tavily_api_key": "tvly-...",
    "max_results": 5
  }
}
  • "provider": "auto" (Default): Automatically uses Brave or Tavily as soon as you paste a brave_api_key or tavily_api_key. If both keys are empty, it falls back to the built-in zero-config search with duckduckgo or sogou.
  • Force a specific search engine: Change "provider" to "brave", "tavily", "duckduckgo", or "sogou".
  • "max_results": Controls how many search results are returned to the agent per query. The default is 5.
  • Provider-native web search: Some AI providers, such as Anthropic Claude and Moonshot / Kimi, also offer their own built-in server-side web search, which Limitcode can use automatically when supported by the active model.
Working with Code

Creating, Editing & Refactoring Files

In Code or Debug mode, Limitcode can make changes directly to your project files once you approve them:

  • Targeted edits: Update specific functions, imports, or classes inside large files without rewriting the rest of the file.
  • Multi-file patches: Apply coordinated diffs across several files during a refactor or feature rollout.
  • New file scaffolding: Create new modules, components, tests, or configuration files in your project tree.
  • Automatic LSP error self-correction: If you use Sublime Text's LSP package, Limitcode checks for fresh syntax and type diagnostics right after editing a file. If the edit introduced a type error or missing import, those diagnostics are fed back to the agent automatically so it can fix the issue on the spot.
Working with Code

Native Diff Review, Undo & Snapshot History

You never have to worry about an AI agent silently breaking your working tree. Limitcode gives you three layers of edit safety:

1. Native Sublime diff view

Proposed changes can be reviewed inside Sublime Text's own diff interface, complete with an inline toolbar to Accept, Reject, or jump to the Next / Prev hunk.

2. One-shortcut Undo & Redo

Before Limitcode modifies any file, it saves an automatic snapshot of the file's state. Press Ctrl+Alt+U or run Limitcode: Undo Last Agent Edit at any time during your session to revert the agent's last change, or press Ctrl+Alt+Shift+R via Redo Last Agent Edit to re-apply it.

3. Full Snapshot History & conflict protection

Want to see every file backup captured during the session? Press Ctrl+Alt+Shift+V or run Limitcode: Show Snapshot History. Limitcode also watches for external edits so the agent won't accidentally overwrite changes you made manually after it read a file.

Working with Code

Running Shell Commands, Builds & Background Jobs

Limitcode can run terminal commands right from the conversation — such as pytest, npm test, cargo check, linters, or git status — and read the output to verify its own work.

  • Approval before execution: Every shell command asks for your confirmation by default so you can inspect the exact command before it runs.
  • Background jobs: Need to start a dev server, file watcher, or long build? Limitcode can launch the command as a background job, check its logs later, or stop it when finished — while you keep chatting and editing.
  • Configurable shell: Choose your preferred shell — pwsh, powershell, bash, zsh, git-bash, or cmd — via the terminal_shell setting.
Control & Autonomy

Work Modes: Scope What the Agent Can Do

Not every prompt should have permission to edit files or run terminal commands. Press Ctrl+Alt+Shift+M or run Limitcode: Change Mode to switch between four specialized modes depending on your current step:

Mode When to use it What the agent can do
Code Everyday feature building, refactoring, and multi-file implementation. Read, search, create, edit, and patch files; run shell commands and background jobs; update the live task checklist; load Skills and MCP tools.
Ask Exploring an unfamiliar repository, understanding logic, or asking questions with zero risk of modifying files. Read-only access: list directories, search across files, read code, and consult Skills or MCP tools. File edits and shell commands are disabled.
Architect Designing a feature, writing technical specs, or planning a complex migration before touching production code. Explore and search the project, read files, and write documentation or plan files without running shell commands.
Debug Tracking down a bug, analyzing stack traces, running reproduction scripts, and applying targeted fixes. Search and read files, run diagnostic commands and test suites, apply targeted code edits or diffs, and verify fixes.
Control & Autonomy

Switching Models & Adjusting Reasoning Effort

Unified Model Selector

Press Ctrl+Alt+M to see every model available across all of your configured providers in one searchable list. Picking a model updates both the provider and model together, and you can switch models mid-conversation without losing your chat history.

Reasoning Effort

Working on a tricky architectural problem or subtle bug? Press Ctrl+Alt+R or run Limitcode: Set Reasoning Effort to choose how much thinking time the model spends before answering: off, low, medium, or high.

Control & Autonomy

Permissions & Safety Boundaries

You decide how much autonomy Limitcode has. Every tool and file pattern can be set to allow to run automatically, ask to prompt you first, or deny to block completely.

Safe defaults out of the box

By default, reading files, listing directories, and searching code run automatically so the agent can inspect your project without pestering you. Any action that mutates files, runs a shell command, accesses the web, or calls an MCP server asks for your approval first.

Session approvals & quick toggles

When prompted, you can approve an action once or for the rest of the current session. Want full speed on a trusted sandbox project? Press Ctrl+Alt+Shift+Y to run Trust Agent — Allow All. Want maximum caution? Press Ctrl+Alt+Shift+U to run Untrust Agent — Ask All.

Protecting secrets & external folders: Use permission.file.deny in settings to block sensitive files (such as .env, **/*.pem, **/*.key), and use outside_project ("ask" or "deny") to control whether the agent can touch files outside your open Sublime workspace.
Control & Autonomy

Delegating Focused Work to Subagents

When a task involves heavy codebase exploration, writing a detailed plan, or running tests alongside an implementation, Limitcode can delegate that piece of work to a specialized subagent. Each subagent works in its own isolated context and reports a clean summary back to your main chat — keeping your primary conversation fast and focused.

Subagent What it is used for Allowed capabilities
explore Map out directory structure and locate where specific logic lives in the codebase. Read files, list folders, search patterns (read-only)
general Investigate a broad question across multiple files and return a concise summary to the main chat. Read files, list folders, search patterns (read-only)
plan Draft a step-by-step implementation plan in Markdown based on your project's actual code. Read files, list folders, search patterns (read-only)
implement Carry out a self-contained coding subtask (creating or editing specific files). Read, list, search, create, edit, and patch files
test Execute your test suite or verification commands and report pass/fail results. Read files, list folders, run shell commands
Control & Autonomy

Long Sessions & Cost Control

Automatic context management

On long refactors or multi-hour debugging sessions, you don't need to manually reset the chat when approaching a model's token limit. Limitcode automatically summarizes earlier turns while preserving critical technical details, files, and decisions so you can keep working seamlessly.

Live token & cost tracking

Press Ctrl+Alt+Shift+C or run Limitcode: Show Cost Statistics anytime to inspect input and output tokens, current context window usage, and estimated session cost based on model pricing. You can also configure an optional budget alert to notify you when a session passes your threshold.

Extending Limitcode

Project Rules: AGENTS.md

Stop repeating your tech stack, naming conventions, and test commands in every prompt. Place an instruction file in your repository root and Limitcode will discover and load it automatically at the start of every task:

  • Supported filenames: AGENTS.md, CLAUDE.md, .limitcode.md, and INSTRUCTIONS.md.
  • To verify which instruction files are active in your workspace, press Ctrl+Alt+F or run Limitcode: Show Instructions from the Command Palette.
Extending Limitcode

Agent Skills: .agents/skills

Agent Skills let you package reusable workflows, domain-specific instructions, code templates, and reference guides that the agent loads on demand only when a task needs them.

  • Project skills: Place skills in .agents/skills/<skill-name>/SKILL.md inside your repository to share them with your team.
  • Global skills: Place skills in ~/.agents/skills/<skill-name>/SKILL.md to make them available across all your Sublime projects.
  • Manage skills from the chat: Press Ctrl+Alt+G to open your skills folder, or simply ask Limitcode in the chat to search for, create, or install a skill for you.
Extending Limitcode

Connecting MCP Servers

Connect external tools and services using the Model Context Protocol (MCP) — such as Playwright for browser testing, database inspectors, issue trackers, or Stripe and GitHub APIs.

To add an MCP server, press Ctrl+Alt+Q or run Limitcode: Open MCP Settings from the Command Palette, then add your local command servers or remote HTTP servers under the "mcp" block:

{
  "mcp": {
    "playwright": {
      "type": "local",
      "command": ["npx", "-y", "@playwright/mcp@latest"],
      "enabled": true
    },
    "stripe": {
      "type": "remote",
      "url": "https://mcp.stripe.com",
      "headers": {
        "Authorization": "Bearer {env:STRIPE_RESTRICTED_KEY}"
      },
      "enabled": true
    }
  }
}

Use {env:VARIABLE_NAME} in headers to reference environment variables without hardcoding secrets in your settings file.

Limitcode Pro using a Playwright MCP server to run end-to-end browser verification from Sublime Text.

Limitcode Pro using a Playwright MCP server to run end-to-end browser verification from Sublime Text. 0:53

Reference

All 16 Agent Capabilities

Here is the complete list of actions Limitcode Pro can perform on your behalf, depending on the active work mode and your permission settings:

Action Category What you can use it for
read_file Project Discovery Read the contents of any file in your workspace (or a specific line range) to understand existing code.
list_files Project Discovery Browse the files and folders inside any directory of your project.
search_files Project Discovery Search for functions, symbols, text, or patterns across your entire repository.
write_to_file Code Editing Create a brand-new file or replace the contents of an existing file.
edit_file Code Editing Make precise, targeted replacements inside an existing file without rewriting the whole file.
apply_diff Code Editing Apply multi-hunk diff patches to a file.
execute_command Terminal & Builds Run shell commands such as test suites, linters, compilers, package managers, or git commands.
background_job Terminal & Builds Monitor the output of, or stop, a long-running command started in the background (like a dev server).
web_search Web & Docs Search the live web (via Brave, Tavily, DuckDuckGo, or Sogou) to check up-to-date documentation or solutions.
web_fetch Web & Docs Open and read a specific documentation URL or web page.
skill Extensions Load specialized instructions, templates, and reference files from an installed Agent Skill.
skill_manager Extensions List, search, create, or install Agent Skills in your project or global skills folder.
mcp Extensions Discover and call tools exposed by your configured Model Context Protocol (MCP) servers.
task Orchestration Delegate a focused subtask (explore, plan, implement, test, or research) to a child subagent.
batch Orchestration Run multiple independent actions (such as reading several files at once) in parallel to save time.
update_checklist Orchestration Create and update the live step-by-step task checklist displayed inside your chat panel.
Reference

Command Palette Reference

Open the Sublime Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and type Limitcode: to access every command:

Chat & Sessions

  • Limitcode: Open Chat
  • Limitcode: New Chat
  • Limitcode: Chat History
  • Limitcode: Rename Session
  • Limitcode: Delete Session
  • Limitcode: Clear Chat
  • Limitcode: Increase Chat Font Size
  • Limitcode: Decrease Chat Font Size
  • Limitcode: Reset Chat Font Size

Code, Models & Autonomy

  • Limitcode: Send to Agent
  • Limitcode: Change Model
  • Limitcode: Change Provider
  • Limitcode: Change Mode
  • Limitcode: Set Reasoning Effort
  • Limitcode: Setup Provider API Key
  • Limitcode: Toggle Permissions
  • Limitcode: Show Instructions
  • Limitcode: Show Cost Statistics
  • Limitcode: Open MCP Settings

Diffs, Undo & Snapshots

  • Limitcode: Undo Last Agent Edit
  • Limitcode: Redo Last Agent Edit
  • Limitcode: Show Snapshot History

Account Sign-In (OAuth & CLI)

  • Limitcode: Login with GitHub Copilot (OAuth)
  • Limitcode: Login with Claude Code
  • Limitcode: Login with Codex (OAuth)
  • Limitcode: Login with Google Antigravity (OAuth)
  • Limitcode: Logout OAuth

Limitcode Pro License

  • Limitcode: Activate License
  • Limitcode: Show License Status
  • Limitcode: Deactivate This Device
Reference

Keyboard Shortcuts

Default keybindings included out of the box (customizable via Preferences → Package Settings → Limitcode → Key Bindings):

Ctrl+Alt+L Open Limitcode Chat
Ctrl+Alt+A Send current selection or active file to the agent
Ctrl+Alt+N Start a new chat session
Ctrl+Alt+H Browse and resume saved project chats
Ctrl+Alt+Y Rename the current chat session
Ctrl+Alt+D Delete a saved chat session
Ctrl+Alt+C Clear the current chat view
Ctrl+Alt+M Change active model (unified selector)
Ctrl+Alt+P Change active provider
Ctrl+Alt+Shift+M Switch work mode (Code, Ask, Architect, Debug)
Ctrl+Alt+R Set model reasoning effort (off / low / medium / high)
Ctrl+Alt+T Show or hide model thinking / reasoning blocks
Ctrl+Alt+U Undo the last file edit made by the agent
Ctrl+Alt+Shift+R Redo the last undone agent edit
Ctrl+Alt+Shift+V Open the file snapshot history
Ctrl+Alt+O Toggle permission prompts
Ctrl+Alt+Shift+Y Trust agent for this session (Allow All)
Ctrl+Alt+Shift+U Untrust agent (Ask All)
Ctrl+Alt+K Configure a provider API key
Ctrl+Alt+S Open Limitcode settings
Ctrl+Alt+Q Open MCP servers configuration
Ctrl+Alt+G Open Agent Skills folder
Ctrl+Alt+F Inspect active repository instructions (AGENTS.md)
Ctrl+Alt+Shift+C Show session token usage and cost statistics
Ctrl+Alt+B Sign in with GitHub Copilot
Ctrl+Alt+E Connect Claude Code
Ctrl+Alt+Z Sign in with OpenAI Codex
Ctrl+Alt+Shift+G Sign in with Google Antigravity
Ctrl+Alt+Shift+L Log out of OAuth account
Ctrl+Alt+X Cancel the active response
Ctrl+Alt+Shift+X / Esc Stop the agent immediately
Enter Send prompt (when chat input is focused)
↑ / ↓ Cycle through your previous prompts in the chat input
Reference

Settings (Limitcode.sublime-settings)

Press Ctrl+Alt+S or go to Preferences → Package Settings → Limitcode → Settings to customize providers, permissions, and appearance:

{
  "default_provider": "openrouter",
  "default_model": "anthropic/claude-sonnet-5",
  "api_keys": {
    "openrouter": "sk-or-...",
    "anthropic": "sk-ant-...",
    "openai": "sk-..."
  },
  "temperature": 0.7,
  "max_tokens": "auto",
  "max_iterations": 50,
  "reasoning_effort": "medium",
  "chat_font_size": "auto",
  "show_thoughts": true,
  "show_cost_estimate": true,
  "web_search": {
    "provider": "auto",
    "brave_api_key": "",
    "tavily_api_key": "",
    "max_results": 5
  },
  "permission": {
    "outside_project": "ask",
    "allowed_external_paths": [],
    "tool": {
      "read_file": "allow",
      "list_files": "allow",
      "search_files": "allow",
      "write_to_file": "ask",
      "edit_file": "ask",
      "apply_diff": "ask",
      "execute_command": "ask",
      "web_search": "ask",
      "web_fetch": "ask",
      "mcp": "ask"
    },
    "file": {
      "allow": ["src/**", "lib/**", "app/**", "*.py", "*.js", "*.ts", "*.md"],
      "deny": ["**/*.lock", ".env", "**/.env*", "**/*.pem", "**/*.key", "**/node_modules/**"]
    }
  }
}
Setting What it controls
default_provider / default_model The active AI provider and model used when opening a chat. Updated automatically when you switch models via Ctrl+Alt+M.
api_keys Dictionary of API keys per provider (e.g. openrouter, anthropic, openai, gemini, deepseek, mistral, groq).
provider_base_urls Custom base URLs per provider (used for self-hosted gateways, proxies, or custom OpenAI-compatible endpoints).
reasoning_effort Controls how deeply reasoning-capable models think before answering: "off", "low", "medium", or "high".
temperature Sampling temperature for generation (e.g. 0.7), or "auto" to use the model's recommended default.
max_tokens Maximum output tokens per response (integer) or "auto" to use the model's native limit.
max_iterations Maximum number of autonomous steps the agent can take in a single turn before pausing.
permission Rules controlling which tools and file patterns run automatically ("allow"), prompt for confirmation ("ask"), or are blocked ("deny").
web_search Configure web search: provider ("auto", "brave", "tavily", "duckduckgo", "sogou"), brave_api_key, tavily_api_key, and max_results. Works out of the box via DuckDuckGo/Sogou, or upgrades to Brave/Tavily when a key is added.
instruction_files List of project instruction filenames to load automatically (AGENTS.md, CLAUDE.md, .limitcode.md, INSTRUCTIONS.md).
auto_discover_instructions When true, automatically finds and loads instruction files in your project hierarchy.
skills / mcp Configuration for Agent Skills and connected MCP servers.
chat_font_size / chat_style Set a dedicated font size for the chat view (or "auto" to match Sublime) and customize visual accents.
show_thoughts / show_cost_estimate Show or hide collapsible model reasoning blocks and live session cost estimates.
skip_dirs / skip_extensions Folders (like node_modules, .git, dist) and file extensions excluded from directory listings and searches.
terminal_shell Preferred shell for running commands (e.g. "pwsh", "powershell", "bash", "zsh", "git-bash", "cmd").
Reference

Troubleshooting & FAQ

Does Limitcode require an external daemon or background service?

No. Limitcode runs directly inside Sublime Text. When you connect an API-key provider, requests go straight from the plugin to your provider with zero background daemons. Close the chat tab and Sublime Text is completely untouched.

How do I pass unsaved code or a specific function to the agent?

Highlight the lines you want to share (or simply focus the file) and press Ctrl+Alt+A (Limitcode: Send to Agent). That attaches your exact selection or buffer contents to the chat. Note that when the agent reads files on its own, it reads the saved files on disk.

Can I review changes before they are saved, or revert an edit I don't like?

Yes. You can review proposed edits in Sublime Text's native diff view (with Accept, Reject, Next, and Prev controls). In addition, Limitcode automatically backs up any file before editing it, so pressing Ctrl+Alt+U (Undo Last Agent Edit) immediately restores the previous version.

Can Limitcode see screenshots and images?

Yes. Copy any image or screenshot to your clipboard and press Ctrl+V (or Cmd+V on macOS) inside the chat view. Vision-capable models can inspect UI bugs, error dialogs, architecture diagrams, or design mockups.

How do I prevent the agent from reading or editing sensitive files like .env?

Add file glob patterns under permission.file.deny in your Limitcode settings (files like .env, *.pem, and *.key are denied by default). Blocked files cannot be read or modified by the agent.

Can I use local models offline with Ollama or LM Studio?

Yes. Select ollama, lm-studio, or litellm as your provider (or configure any custom OpenAI-compatible base URL). Your prompts and code stay on your local machine.

How do I move my Limitcode Pro license to a new computer?

Your license supports up to 3 personal devices at once. To free up a seat on an old machine, open the Command Palette and run Limitcode Pro: Deactivate This Device, then activate on your new computer.

What happens if I lose internet access while traveling?

Once your device is activated, Limitcode Pro caches a signed offline pass that lets you continue working offline for up to 30 days between license checks.

Direct support from the creator

Need help setting up your workflow?

Ask about installation, licensing, or connecting your preferred AI provider — in English or Spanish.

Get Limitcode Pro — $29 → Email hola@jawuil.dev →