Brain
Brain is the long-term memory layer. Anything worth keeping across sessions lives here: notes you write, memories the agent saved, files you uploaded, and signed agreements. The Brain tab in the right panel is your view onto it. The chat turn that runs five minutes later automatically retrieves the right items so you do not have to remind the agent of context.
Overview
- Right panel tab
- Brain (icon: brain glyph)
- Surfaces
- Notes, Memories, Files, Agreements
- Search
- Full-text plus semantic across all four
- Retrieval in chat
- Automatic; hybrid semantic and recency
- Share
- Notes can be turned into read-only share links
- Write paths
- Manual (you), agent-saved (skill runs), auto-summary (every 5 chat messages after the 10th)
- Edit
- Notes are editable; Memories are read-only after creation
Brain is the part of the platform that turns repeated conversations into actual leverage. Every meeting transcript, every research result, every decision you wrote down is one search away from the agent. The agent does not have to ask you who the prospect is, what your offer is, or what tone you write in. It already knows.
Four surfaces
Each card on the Brain tab is one kind of content. Pick the one that matches how the data should behave.
| Surface | Created by | Editable | Surfaces in chat as |
|---|---|---|---|
| Notes | You (manual write or promoted memory) | Yes | Inline note block in the agent reply when relevant |
| Memories | Agent (skill runs and auto-summary) | No | Stitched into the system prompt for the turn |
| Files | You (drag and drop or paste) | No (replace instead) | Referenced by name when the agent needs the content |
| Agreements | Signing flow (contracts, NDAs) | Status only | Linked when the agent works on a related deal |
Notes
Human-written prose you want to keep, edit, and reuse.
Notes are the only Brain surface you create directly. Use them for things you want to revisit yourself: meeting notes, decisions, brand positioning paragraphs, message templates, anything that needs to be edited later.
- 01Create a noteClick the plus button on the Brain tab. The note opens in an editor with a title field and a body. Save persists immediately.
- 02Promote a memory to a noteOpen any memory in the detail view. The header has a Move to Notes action. It copies the memory's text into a new editable note and tags it accordingly.
- 03Edit and re-saveClick any note to open the editor. Changes are saved on blur or on explicit save. Old versions are not kept; this surface is meant to be living.
Memories
Agent-written, retrieval-indexed, read-only after creation.
Memories are short structured records the agent writes after running a skill or finishing a turn that produced something worth keeping. Each memory has an owning agent, free-text content, and a set of tags that describe what it is.
You do not edit memories. If a memory is wrong or stale, you delete it. If you want a memory you can edit, promote it to a note first.
| Memory kind (tag) | Created when |
|---|---|
| session-summary | Auto-summary after 10+ chat messages, every 5 messages from then on |
| working | Skill saves a working note about an in-flight task |
| knowledge | Skill saves a long-term learning, research result, or fact |
| profile | Skill saves a fact about you or your company (brand voice, ICP, pricing model) |
| skill-run | Marks the completion of a named skill execution; surfaces in the activity feed |
Files
Documents you uploaded. The agent can read and reference them.
Files are anything you dragged into the composer or attached to a turn. PDFs, images, audio, office documents, spreadsheets, markdown. The Brain tab lists them by upload date with their original filename. Clicking a file opens a viewer.
Files are referenced by the agent when relevant. You do not have to re-attach the same file in every turn that needs it. A reference like "the brand guide PDF I uploaded last week" works because the retrieval layer matches the description to the file metadata.
Agreements
Contracts and NDAs that have been sent or signed through Ultron.
Agreements are documents created by a contract drafting skill or uploaded directly. Each agreement shows its status (draft, sent, partially signed, fully signed) and the list of signers with their last action timestamps.
Once an agreement has been signed, the signed PDF lives in the agreement record forever. Re-running the contract drafting skill on the same deal creates a new agreement; it does not modify the signed one.
Search
One bar across all four surfaces.
The Brain tab has a search field at the top. Typing filters all four surfaces at once. Results are grouped by surface and ranked by a mix of semantic match, exact-text match, and recency. The same search bar is what the agent uses internally for memory lookups.
How retrieval works in chat
What the agent sees automatically before every turn.
Before processing your message, the platform queries Brain for items relevant to what you just said. The top few results are stitched into the system prompt for that turn. The agent reads them as context, not as user input, so they do not appear in the visible conversation.
Retrieval is hybrid. Three signals combine:
| Signal | What it catches |
|---|---|
| Semantic similarity | Memories about the same topic even if the words differ. 'Pricing strategy' surfaces a memory titled 'How we package the offer'. |
| Tag and agent filter | Memories owned by the agent you are talking to, or tagged for the kind of work you are doing. |
| Recency | Recently created memories get a boost so what you said yesterday outranks what you said a year ago. |