Skip to main content

Assistant

Turn the Console assistant on or off, add the organisation's OpenRouter key, choose what it may send, switch capabilities and watchers, and allow AI apps.
Open it fromOrganisation settingsAssistantRoles:OwnerPlatform administratorConsole address: /org-settings/assistant

The Assistant page holds the organisation's settings for the Console assistant, the helper that answers questions about the page a member has open and drafts control changes for them to review. Each setting is a row on one card, in this order:

  • Assistant in search
  • Model access
  • Data sent to the model
  • Capabilities
  • Watchers
  • MCP server
  • Usage

A change to a switch saves as you make it, with “Saving…” and then “Saved” beside it.

What the assistant sends to the model, who can keep it and what DALP masks first is explained on What the assistant sends. This page covers the controls.

The “Assistant” page in “Organisation settings”.

The Assistant in search switch comes first. “Model access” names the key in use, the free assistant or the organisation's own OpenRouter key, with a note on where questions go. “Data sent to the model” offers “Mask” and “Send documents” with a “What is sent” link. “Capabilities” lists six switches, followed by “Watchers”, “MCP server” and “Usage”.

Steps

Skip the section on adding the organisation's key if the organisation uses the free assistant.

Turn on the assistant

  1. Open “Organisation settings”, then Assistant.
  2. Turn on Assistant in search. While it is off, the row reads “Nobody in this organisation sees the assistant.” and the assistant sends no question to a model.

Add the organisation's key

  1. In “Model access”, select Add your own key.
  2. In OpenRouter key, enter the full key. The field takes sk-or- followed by 8 to 250 printable characters, without spaces, and the form says “Enter a full OpenRouter key: sk-or- followed by at least 8 characters, without spaces.” until the key fits.
  3. Select Save key.

DALP stores the key encrypted and shows “Stored encrypted. After saving, only the last 3 characters are shown.” The row then reads “Key”, “OpenRouter key ending” with the last three characters, and the “Model” the key uses. Select Replace to store a new key in its place. The field is then labelled “New OpenRouter key”.

Choose what the model receives

  1. In “Data sent to the model”, choose Mask to replace holder names, emails and wallet labels with handles and send no documents or memos.
  2. Or choose Send documents to send the document pages and memos the assistant reads as written, including any names inside them. The Console asks “Send documents to the model?” and “Document pages and memos the assistant reads will leave DALP and go to OpenRouter and the model provider it routes to. Names inside them are not masked.” Select “Allow sending” to confirm.

On the free assistant, with no key stored, document pages and memos are sent by default. Choose “Mask” to keep them in DALP. Select What is sent to see what is always sent, what is sent unless you choose “Mask”, and what is never sent. “Read the details” opens What the assistant sends.

Turn the assistant off

Nobody in the organisation sees the assistant after you turn off Assistant in search, and the assistant sends no question to a model. A connected AI app sends what it reads to its own model provider, which “MCP server” controls. The other settings stay as they were.

Remove the organisation's key

  1. In “Model access”, select Remove.
  2. Read the confirmation, then select Remove key, or “Keep key” to stop. The text is “Remove the OpenRouter key? Everyone moves to the free assistant.” Where the installation has no free assistant it is “Remove the OpenRouter key? Nobody can use the assistant until a key is added again.”
  3. When documents are being sent and the installation has a free assistant, the confirmation adds “Documents will still be sent, through the free assistant. Choose Mask to stop that.”

Removing the key leaves the Data sent to the model choice as it was.

Keep only some capabilities on

Each capability under Capabilities has its own switch, and all six are on until an owner or a platform administrator turns one off. The page says “Turn a capability off and nobody in this organisation sees it. The assistant switch above turns them all off at once.”

  • Answers: questions about the asset, participant, trading venue or audit trail you are looking at.
  • Drafts: a control change prepared from a request or a pasted text, for you to review and sign.
  • Term sheet check: compares a term sheet with the terms on the chain and fills “Issue asset” from one. It needs documents allowed.
  • Holder notices: the wording of a notice to holders for a corporate action.
  • Approval checks: the brief and the one-line check on a pending approval.
  • Typed filters: a phrase typed above a table, turned into filters.

With “Answers” off and Drafts on, only an asset page offers the assistant, for drafts, and every other page shows plain search. “Drafts” also covers reading a written compliance policy into rules, on the Compliance tab of an asset and in the Asset designer, so that reading is refused while “Drafts” is off. A capability that is off is refused on the server as well, so a hidden screen is not the only guard.

Switch a watcher on

A watcher reads DALP data on a schedule and posts what needs attention to the Actions inbox of the people whose roles it concerns. Every watcher stays off until you switch it on. Watchers lists those of the four watchers this page lists that your roles concern, or “No watchers for your roles” when there are none:

  • Payout funding: every hour, checks that each treasury can pay the dividends, distributions and coupons due in the next 14 days, and prepares the top-up when it cannot.
  • Filing deadlines: every six hours, finds filings due in the next 14 days that have no submission.
  • Verification expiry: every six hours, finds holder verifications that an asset's compliance rules require and that expire within 30 days.
  • Unmatched credits: every hour, finds bank credits that match no client after 24 hours and proposes what to do with each.

Turn on the watcher's switch. Each row then shows “Last run”, “Next run” (“Off” while the watcher is off) and “Open items”.

Select Run now to request a run, which then reads “Run requested”.

Show activity lists the latest runs with what each “Found”, “Posted”, “Cleared” and “Executed”, and “Hide activity” closes the list. A watcher sends nothing to the model, so it needs only the assistant switched on, not a key.

Set a watcher's level

Each watcher shows its “Level”, its “Track record” and what is “Needed to raise” it. Choose another option in Level. Lowering a level takes effect at once, and the next run uses it. An option the watcher cannot take is unavailable, and the line under “Level” says why.

  • Observe: runs and logs what it finds. Posts nothing.
  • Suggest: posts what it finds to the Actions inbox, without a draft.
  • Prepare: posts what it finds with a prepared draft or options for a member to take. A new watcher starts here, or lower when it cannot prepare anything.
  • Act with veto window: schedules the action under the organisation's agent principal and carries it out unless a member vetoes it within the window.
  • Act: the platform carries out the action with the agent principal's platform-held key, inside its on-chain mandate, with no member's signature, and reports it.

An agent principal is an account the platform signs for on the organisation's behalf, limited by an on-chain mandate. The two act levels stay out of reach until the organisation has an active agent principal. The platform checks the agent's mandate when the action runs and refuses an action the mandate does not cover. Only the organisation owner raises a level. Raising it to “Prepare” or higher also needs the watcher's track record to meet the minimum. The owner sets that minimum, down to 0 items and 0 percent, and can change it in the same request as the raise, so the track record does not hold back an owner.

Needed to raise shows the minimum as “{n} resolved in 90 days, {n}% taken unchanged”, and the owner edits it with “Change”, “Resolved items” and “Taken unchanged (%)”, then “Save” or “Cancel”.

Track record reads “Last 30 days” and “Last 90 days” with the proposals taken, changed and rejected, or “no proposals resolved”. The reasons the line gives include “Only the organisation owner can raise this watcher to Act.” and “The last 90 days fall short of the track record needed for Act.”

Only Payout funding has an action: the transfer of a treasury's shortfall in the payout asset. At “Act with veto window” it waits four hours after the item is posted, and anyone who sees the item, an organisation owner or a holder of the watcher's system role, can select “Veto transfer” on it in the Actions page before then. A veto closes the item as rejected. A transfer cannot be undone once the agent principal has run it.

Backtest replays a watcher's checks once a day over indexed history and says how often it would have fired, as “Would have fired {n} times in the last 30 days”, without posting or changing anything. “Hide backtest” closes it.

Stop all watchers drops every watcher to “Observe” at once, after you confirm with “Stop all watchers” or choose “Keep them running”.

Connect AI apps through MCP

MCP server lets members connect their own AI apps, such as Claude or ChatGPT, to DALP through the Model Context Protocol. An app reads what the member can read and prepares actions in the Actions inbox for the member to sign. An app with “Prepare actions” access can also run a mint, burn, pause, resume, block or unblock from the account of one of the organisation's agent principals, inside that agent's on-chain mandate, with no signature. The member's own wallet must also hold the permission for that action on the asset. Turn the switch on. While it is off the row reads “Off. Members cannot connect apps to DALP.” A person who cannot change it sees “Only an owner can change this setting.”

While the assistant is on, Connected apps lists each connection with these columns:

  • App
  • Member
  • Access, which is “Read” or “Prepare actions”
  • Connected
  • Last used, which reads “Never” before first use

Select “Revoke” on a row, then Revoke access. For an owner the confirmation says “The app loses access for every member who connected it. Members can connect it again.”

Read the usage

Usage shows the questions asked in the last 7 days, the questions refused because the assistant was busy, the limits that apply, and the answers rated helpful and not helpful. It reads “No questions in the last 7 days.” when there are none. “See these questions in the audit trail” opens them for a reader who can open the Audit trail. The row appears only to an owner or a platform administrator. What the assistant sends lists the limits.

Troubleshooting

What you see
What it means and what to do
“Assistant settings unavailable” with “Nothing was changed. Try again.”
The settings could not be read. Reload the page and try again. Usage and Watchers report their own failures, “Usage is unavailable.” and “The watchers could not be read.”, without blocking the other settings.
Nobody in the organisation sees the assistant
Check that Assistant in search is on, that Model access names the free assistant or shows a key, and that the capability you expect is on under Capabilities. With no free assistant and no key, add a key first.
“The key could not be saved.” or “Enter a full OpenRouter key: sk-or- followed by at least 8 characters, without spaces.”
Copy the whole key again. It must start with sk-or- and contain no spaces.
Send documents is unavailable, with “Only an organisation owner can allow this.”
The person cannot change the settings. Ask an owner or a platform administrator to choose it.
“The assistant setting could not be saved.” or “The data setting could not be saved.”
The change was not stored. Try the switch again.

API routes

The page reads and changes its settings through GET /api/v2/settings/assistant and PATCH /api/v2/settings/assistant. Watchers use /api/v2/assistant/watchers, with /api/v2/assistant/watchers/{watcherId} to switch one on or change its level, /api/v2/assistant/watchers/{watcherId}/runs to list and request runs, /api/v2/assistant/watchers/{watcherId}/backtest for the backtest and POST /api/v2/assistant/watcher-stops to stop them all. Connected apps are at /api/v2/assistant/mcp-clients and DELETE /api/v2/assistant/mcp-clients/{clientId}. Request and response fields are in the v2 API reference.

Was this page helpful?