# Commands and settings

> Every command, chat shortcut and setting, with defaults, and which settings a repository is not allowed to change.

GBS SE Agent · Reference · Source: https://kit.gramproindia.com/se-agent/commands-and-settings · Code: https://github.com/gramproindia/agent

## Chat commands

| Type | What it does |
| --- | --- |
| `/new` | Start a new chat |
| `/compact` | Summarise the conversation to free context |
| `/skills` | List skills, their source and what they cost per request |
| `/mcp` | List MCP servers, their state and their tools |
| `/help` | Show the commands |
| `@` | Attach a file or folder to the message |
| <kbd>Esc</kbd> | Stop the running agent |

## Command Palette

Open it with <kbd>Ctrl</kbd>+<kbd>Shift</kbd>+<kbd>P</kbd>.

| Command | What it does |
| --- | --- |
| **GBS Agent: New Chat** | Starts a fresh conversation |
| **GBS Agent: Stop Agent** | Stops the running agent |
| **GBS Agent: Set API Key** | Stores a key in secret storage, bound to the current endpoint |
| **GBS Agent: Clear Chat History** | Deletes all stored chats for this workspace |
| **GBS Agent: New Skill** | Scaffolds a skill folder and opens it |
| **GBS Agent: Open Skills Folder** | Opens your personal skills folder |
| **GBS Agent: Reload Skills** | Re-reads skills and asks again about repository skills |
| **GBS Agent: Show MCP Servers** | Lists MCP servers, their state and their tools |
| **GBS Agent: Reload MCP Servers** | Restarts MCP servers and asks again about repository servers |
| **GBS Agent: Show Logs** | Opens the log: requests, token usage, tool calls, errors |

## Settings

| Setting | Default | Purpose |
| --- | --- | --- |
| `gbsAgent.provider` | `ollama` | `ollama` or `openai` (any OpenAI-compatible server) |
| `gbsAgent.model` | `deepseek-v4-flash:cloud` | The main agent's model |
| `gbsAgent.subagentModel` | *(empty)* | A smaller model for exploration subagents |
| `gbsAgent.permissionMode` | `acceptEdits` | `ask`, `acceptEdits` (Auto-edit) or `auto` (Full auto) |
| `gbsAgent.ollama.baseUrl` | `http://localhost:11434` | The Ollama server |
| `gbsAgent.openai.baseUrl` | `https://router.huggingface.co/v1` | The OpenAI-compatible server, local or cloud |
| `gbsAgent.contextWindow` | `32768` | The model's context size; sent to Ollama as `num_ctx` and used to decide when to compact |
| `gbsAgent.maxOutputTokens` | `8192` | Maximum tokens per response |
| `gbsAgent.temperature` | `0.2` | Sampling temperature |
| `gbsAgent.maxSteps` | `60` | Model calls per request before the agent pauses |
| `gbsAgent.commandTimeoutSeconds` | `120` | Default timeout for shell commands and MCP tool calls |
| `gbsAgent.maxToolOutputChars` | `20000` | Tool results longer than this are trimmed before the model sees them |
| `gbsAgent.shell` | *(empty)* | Shell for commands (default: cmd.exe on Windows, `$SHELL` elsewhere) |
| `gbsAgent.skills.enabled` | `true` | Use skills at all |
| `gbsAgent.skillsPath` | *(empty)* | Your personal skills folder (default: the extension's own folder) |
| `gbsAgent.mcp.enabled` | `true` | Start MCP servers and give their tools to the agent |
| `gbsAgent.mcpServers` | `{}` | MCP servers for every project, in `.mcp.json` format |

### Settings a repository cannot change

The provider, model, subagent model, endpoints, API key settings, permission
mode, shell, skills path and MCP servers are read from your **User** settings
only. A project's `.vscode/settings.json` cannot redirect your code to another
server, run a different shell, or switch off approvals.

API keys belong in secret storage, through **GBS Agent: Set API Key**. The
older `gbsAgent.openai.apiKey` and `gbsAgent.huggingfaceToken` settings still
work but are deprecated, because they store the key as plain text.

## Fixed limits

Some limits are part of the code rather than settings, because the right value
does not depend on your model:

| Limit | Value |
| --- | --- |
| Lines read from a file at once | 1,000 (the model can ask for a range) |
| Subagents running at the same time | 4 |
| Subagents per request | 12 |
| Command output kept | the first 6,000 and last 14,000 characters |
| A stream that sends nothing | aborted after 5 minutes |
| Retries on 429 or a gateway 5xx | 3, honouring `Retry-After` up to a minute |
| Skills listed in the prompt | 12 |
| A skill's instructions / one reference file | 6,000 / 12,000 characters |