AI agent page: Sen, the broker assistant and notes
The AI page holds three tabs — the market agent Sen, a broker-connection assistant and a notes panel — with a conversation rail on the left, a sources panel on the right, and a chat that streams its answer token by token.
Senzoukria · Documentation · Updated September 2026
Where to find it
- Where
- Left navigation → AI
- Tabs
- Sen (market agent), broker access assistant, Notes
- Attachments
- 4 images maximum per message — PNG, JPEG, GIF or WebP
- Send
- Enter sends; ⇧↵ inserts a newline
- Access
- A Claude key, a Groq key, or a reachable local engine that declares at least one model
What it does
Sen reads what is on your screen — flow, delta, heatmap, gamma — and answers about the current session. The empty state states the boundary in the product's own words: it does not predict and does not give financial advice, it reads what is there. The broker access tab is a separate conversation with its own system prompt, built on what the connection diagnostic measured; the two are never mixed, because an agent discussing delta and gateways in one thread no longer knows which context wins.
Only one chat is mounted at a time. The conversation itself lives in a store rather than in the component, so switching tabs loses nothing. Replies stream in: text arrives as deltas into the assistant bubble, and the bubble is finished only when the model stops for real — a turn that pauses to call a tool keeps writing under the same request id, so tool use never looks like a truncated answer. Token counts are attached to the bubble when it completes.
Controls
| Setting | Default | What it changes |
|---|---|---|
| Sen / broker access / Notes tabs | Sen | Which conversation or panel is mounted; only one at a time |
| Threads rail | Open | Collapses or expands the saved-conversation rail; locked while a reply is being written |
| Sources panel | Open on screens ≥ 1000 px | Shows the sources the agent read; hidden in focus mode |
| ⧉ Detach | — | Opens the assistant in its own window |
| ⚙ Settings | — | Deep link to the AI section of the settings page |
| Attach a screenshot | None | Adds up to 4 images to the next message; refused beyond that with "4 images maximum per message." |
| New conversation | — | Starts a fresh thread; the previous one was already saved when its last reply ended |
How to use it
- Ask about the session in front of you. The suggestion chips give the shape of a good question: explain absorption with a diagram, read today's GEX key levels when an options source is configured, which indicators help read a reversal, which economic events shape the session.
- Attach a screenshot when the question is about something visible. Images are placed before your text in the turn, so the model reads the question already knowing what it is looking at. Only the Claude engine reads images — Groq and local models receive your text alone.
- Open a saved thread from the rail to pour it back into the chat. A reopened thread has no archived tool trace, and none is invented for it.
- If you get stuck during a broker setup, the button in the header switches you straight to the connection assistant with your step pre-filled.
Limits and pitfalls
- App tools are read-only: the agent can read the chart, the broker account, the calendar and news, backtests, scripts and the autopilot state — never place an order and never change a script.
- Answers are billed by whichever provider you chose, on your own key. Market data itself is billed by your data provider, not by the assistant.
- The agent describes what it reads. It gives no financial advice, no prediction and no promise of a result.
- The same chat component is also mounted on the welcome dashboard card. They share one store, so they are kept on mutually exclusive routes — do not keep both mounted at once.
Related pages
This page in other languages
Frequently asked questions
- Do I need a paid key to use the assistant?
- No. The page opens as soon as there is something able to answer: a Claude key, a Groq key, or a reachable local engine that declares at least one installed model.
- Why did my screenshot not reach the model?
- Only the Claude engine reads images. On Groq or a local engine the request carries plain text, and the picker says so before you attach anything.
- Can Sen place a trade for me?
- No. Its tools read the application; they never send an order and never modify a script. Order entry stays a deliberate human action, and the autopilot is armed by hand.