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 |
| Esc | Stop the running agent |
Command Palette
Open it with Ctrl+Shift+P.
| 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 |