OnlySq CLI
AI coding agent for VS Code
Chat · Agent mode · Inline completions · MCP · Semantic search
✨ Features
💬 Chat
- Stream responses from any OnlySq-hosted model — GPT, Claude, Gemini, Grok, DeepSeek, Qwen and more
- Three modes: Agent ⚡ (autonomous tool use), Chat 💬 (conversation), Plan 📋 (read-only analysis)
- Model badge on every response — always see which model answered
- @-mentions — type
@filename to attach workspace files as context
- Drag & drop files and images directly into the composer
- Multi-chat with search, rename, and persistent history
- Export any conversation as Markdown or JSON
- Slash commands —
/model, /memory, /export, /clear, /help
🤖 Agent
Autonomous coding agent with 37 built-in tools and up to 500 steps per run.
| Category |
Tools |
| File ops |
read_file, propose_edit, replace_in_file, apply_at_line, patch_file, delete_file, rename_file |
| Search |
search (regex), semantic_search (embeddings), find_files (glob), find_in_file |
| Shell |
run_command, run_command_interactive, terminal (persistent sessions) |
| Git |
git_status, git_diff, git_commit |
| Web |
fetch_url, web_search, scrape_page, open_in_browser |
| Editor |
open_file, goto_position, get_cursor, get_selection, get_diagnostics |
| VS Code |
run_vscode_command, run_task, list_tasks |
| Memory |
add_memory, get_memory, view_memories, delete_memory |
| Tasks |
create_task, update_task, delete_task, list_agent_tasks |
| Context |
read_project_context, update_project_context, workspace_info, system_info |
| Control |
pause_agent, ask_user, delegate (to sub-agents) |
Live diff preview — every file edit opens a native VS Code diff with Apply / Reject buttons. Batch operations with Apply All / Reject All and Undo.
Cancel button — kill hanging run_command or terminal operations mid-execution.
🧠 Sub-Agents
The agent can delegate tasks to 8 specialized sub-agents that run in isolated contexts:
| Sub-agent |
Purpose |
code_reviewer |
Find bugs and security issues (read-only) |
code_writer |
Implement code changes |
test_runner |
Run and fix tests |
explorer |
Explore and explain codebase structure (read-only) |
shell_operator |
DevOps and system tasks |
refactorer |
Restructure code preserving behavior |
doc_writer |
Write documentation |
planner |
Break complex goals into steps (read-only) |
🔌 MCP (Model Context Protocol)
Connect external tool servers via the MCP standard. Any MCP server works — GitHub, Postgres, Slack, filesystem, and hundreds more.
// .onlysq/mcp.json
{
"servers": {
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..." }
}
}
}
- UI panel in chat header — add, remove, enable/disable, restart servers
- Hot reload — edit
mcp.json and servers restart automatically
- Commands:
OnlySq: MCP — Edit Config, MCP — Reload Servers, MCP — Status
🔍 Semantic Search
Natural-language search across your entire workspace using embedding vectors.
OnlySq: Reindex Workspace — scans files, chunks text, generates embeddings- Agent uses
semantic_search tool — finds code by meaning, not exact keywords
- Incremental updates — unchanged files are skipped on re-index
Configurable embedding model: gemini-embedding-001 (default), gemini-embedding-2, pplx-embed-*.
⌨️ Inline Completions
AI-powered autocomplete in any language. Triggered as you type or on demand.
🖥️ Live Terminal Sessions
Persistent shell sessions shared between you and the agent:
- Agent opens a terminal → you see it in VS Code
- You type in the terminal → agent sees your input on its next step
- Full
open / write / read / peek / close / list lifecycle
💾 Persistent Memory & Tasks
- Memory — key-value store that survives across sessions. Agent remembers your preferences, SSH keys, project decisions.
- Tasks — trackable work items with status (todo → in progress → done). Displayed as compact markers in chat, managed via dedicated panel.
- Project context —
.onlysq/context.md auto-injected into every agent run.
🚀 Quick Start
- Install the extension
- Click the OnlySq icon in the Activity Bar
- Click Auth with OnlySq → sign in via browser
- Start chatting — toggle Agent mode with the ⚡ button
⚙️ Configuration
| Setting |
Default |
Description |
onlysq.chatModel |
gpt-4o-mini |
Model for chat and agent |
onlysq.completionModel |
gpt-4o-mini |
Model for inline completions |
onlysq.embeddingModel |
gemini-embedding-001 |
Model for semantic search index |
onlysq.temperature |
0.3 |
LLM temperature (0–2) |
onlysq.agent.maxSteps |
50 |
Max agent loop iterations (1–300) |
onlysq.agent.parallelTools |
true |
Run independent tools in parallel |
onlysq.agent.toolCache |
true |
Cache read-only tool results within a run |
onlysq.agent.toolPolicy |
{} |
Per-tool approval: always / ask / never / disabled |
onlysq.agent.customSystemPrompt |
"" |
Custom instructions appended to system prompt |
onlysq.agent.personalization |
false |
Let agent learn your style from conversations |
onlysq.chat.persistHistory |
true |
Persist chat history between sessions |
onlysq.inlineCompletions.enabled |
true |
Enable inline completions |
Full list: Ctrl+, → search "OnlySq".
🎹 Keyboard Shortcuts
| Shortcut |
Action |
Ctrl+Alt+O |
Open chat |
Ctrl+Alt+N |
New chat |
| Right-click → Explain Selection |
Explain selected code |
| Right-click → Refactor Selection |
Refactor selected code |
📋 Commands
Open the Command Palette (Ctrl+Shift+P) and type OnlySq:
| Command |
Description |
| Open Chat |
Focus the chat panel |
| New Chat |
Start a fresh conversation |
| Run Agent… |
Prompt the agent with a goal |
| Select Chat Model |
Pick from available models |
| Select Completion Model |
Pick completion model |
| Open Settings |
Open in-chat settings panel |
| Initialize Project Context |
Auto-generate .onlysq/context.md |
| Reindex Workspace |
Build semantic search index |
| Drop Semantic Index |
Delete the index |
| MCP — Edit Config |
Open .onlysq/mcp.json |
| MCP — Reload Servers |
Restart all MCP servers |
| MCP — Status |
Show server status in log |
| Show Log |
Open extension output channel |
| Reset Token Usage |
Reset session token counter |
🗂️ Project Files
| File |
Purpose |
.onlysq/context.md |
Persistent project context, auto-injected into every agent run |
.onlysq/mcp.json |
MCP server configuration |
.onlysq/index.json |
Semantic search index (auto-generated) |
.onlysq / .onlysq-rules / .onlysq.md |
Project rules, auto-read as custom instructions |
Requirements
License
MIT
| |
