SensAI-Agent

SensAI in VS Code, Cursor, or Windsurf. Same account, proxy, and credits as the CLI. The editor never holds a provider API key.

This is the optional extension. The full gold cockpit is SensAI IDE.

Sign in

Command Palette → SensAI: Sign In. The extension opens the SensAI login page in your browser (https://sensai.immunisense.com/sensai-login). Finish there. No CLI command is required for login.

Agent engine

SensAI: Connect Engine starts the ACP engine for the open folder. In SensAI IDE that binary is bundled. In VS Code / Cursor / Windsurf, install the CLI and keep sensai-cli on PATH, or set sensai.engine.path:

irm https://sensai.immunisense.com/install.ps1 | iex

Sense Chat

Open the Sense Chat panel in the secondary sidebar (Ctrl+Alt+L).

• Workflows: Code (tools + permission prompts), Plan (spec first), Analyze (read-only), Chat (no tools), Security (read-only vulnerability hunt; included with Sense, Sense Pro, and Sense Ultra).
• Autopilot: on, SensAI edits files and runs commands without asking; off, every write and shell call shows an Allow / Allow for session / Deny card. Saved per workspace.
• Model, effort, and Sense mode: pick the model and reasoning effort in the composer. Models that offer Sense mode (the TUI's /sense: the full context window at higher pricing, 200K context · 500K with Sense) get a Sense mode switch at the top of the model menu; the model button shows a Sense badge while it is on and the context ring is measured against the larger window. Saved per workspace.
• Streaming: replies stream in with a collapsible Thinking block, a tool timeline in the order things happened (rules, skills, and agents pulled into context fold into one Included rules / Included skills row), and sub-agent cards.
• Sessions: tabs per conversation, restored when you reopen the workspace. + starts a new one; the list icon opens every stored session of the workspace (TUI ones included). The keyboard icon next to it opens the shortcut reference in Settings.
• Keyboard shortcuts: fixed. Enter sends, Shift+Enter inserts a line, Esc stops the turn. The full list is SensAI: Keyboard Shortcuts in Settings. It is a reference; the shortcuts cannot be changed.
• Context: type @ to mention a file or folder (@repo:<name>/… from another workspace folder), #<name> to focus a folder, the paperclip to pick a file, or right-click → SensAI → Add Selection as Context (Ctrl+Alt+K). The ring on the left shows context usage, also for sessions you come back to.
• Checkpoints: prompts whose turn changed files get a Checkpoint · Restore row. Restore reverts the workspace to before that prompt and puts the prompt back in the composer.
• Message actions: copy a reply, or use … for Copy as Markdown, Copy conversation ID, Copy request IDs. Each reply ends with Est. credits used: N.NN · Elapsed time: 2m 17s.
• Stop: press Esc or the stop button while a turn runs.

Autocomplete

Inline suggestions from SensAI while you type; press Tab to accept. Toggle with the ✓ Autocomplete status bar item, SensAI: Toggle Autocomplete, or sensai.autocomplete.enabled. Suggestions use credits like any other request.

Cockpit

The SensAI view in the primary side bar lists the engine, rules and context files, agents, skills, MCP servers, and hooks, each split into Workspace (<repo>/.sensai/…) and Global (~/.sensai/…). Press + (in the title or on a folder) to add one; rules, agents, and skills open as starter files, MCP servers and hooks open the configuration panel.

• MCP status on every server row: Connected (4 tools), Connecting…, Disabled, Connection failed · reason with Retry, or Unauthenticated with Authenticate. A connected server expands to its tools.
• Enable / disable any rule, agent, skill, MCP server, or hook inline, or a whole Workspace / Global folder at once with its Enable all / Disable all icons; disabled items are muted. The setting is shared with the CLI and TUI.
• Configuration panel (SensAI: Configure MCP Servers & Hooks): add and edit servers (stdio / http / sse, command, URL, environment, headers, timeouts, tool allow-lists; secrets are sealed into the OS keyring) and hooks (event, tool types or file patterns, run a command / ask an agent / notify) without touching config.toml.
• Profiles (the TUI's /profile): the panel's Profiles tab sets a default model and effort, one per workflow (Coding, Planning, Research, Analysis, Review), and pins models to sub-agents. Pick a profile in the composer next to Autopilot; the model and effort pickers step aside and every turn and sub-agent follows it. The reply footer shows what the sub-agents spent.

Plans

The Plans view lists the plans in <workspace>/.sensai/plans with their phase. New Plan opens a Plan Mode session that walks Requirements → Design → Tasks; approve each phase and it is written as 01-requirements.md, 02-design.md, 03-tasks.md next to a plan.json the CLI and TUI can pick up.

Multi-repo workspaces

A workspace with several folders is a multi-repo task. The extension keeps <first folder>/.sensai/workspace.toml in step so the engine can open every folder. In Sense Chat, @repo:<name>/path adds a file from another folder as context, and #<name> points the agent at a folder without loading its files.

Right-click in the editor for Explain Selection and Review File.

All model traffic goes through https://sensai.immunisense.com.

Commands

Settings

Issues

Bugs and features: github.com/immunisense/sensai with the extension label. Security reports: security@immunisense.com.

License

Proprietary. See LICENSE.md. IMMUNISENSE CORP.