Skip to content
Optima docs
What's new

What's new in Optima v1.1.0

Optima Agents, consent statistics, clearer schedules and many fixes across the console and the API.

Released 30 October 2026.

Optima Agents

Agents are background AI workers that act on your data with your own model and budget. You write what an agent should do in plain sentences, choose when it runs and which tools it may use, and decide what it may do with the result: email people, record events or set attributes. Start with Optima Agents, or follow the guide Get a weekly digest from an agent.

  • Definitions and instances. A definition describes the work: the prompt, its triggers (an event or a schedule), its tools, its actions, a purpose and parameters. Every save is a new version. An instance runs one version with a model from AI models and a token budget, or a budget in money when the model has prices.
  • Dry run first. An instance starts in dry run, where it shows what it would have done without sending or writing anything. Switch it to live when you are happy with it.
  • Run limits. Each instance has limits on turns, tool calls, run time and how much it may read, set when you create it. The defaults suit most agents.
  • Run log. The run log lists what every run did, including the subject and text of each email, and a short summary of the run in the model's words. Personal data above your clearance is masked.
  • Continuous agents. A continuous agent works through a whole segment or event feed, one profile or event at a time, and then keeps up with changes. Before it starts, a cost preview estimates the tokens it will use. Going live after a dry run processes the dry-run items again, live. See Score customers with a continuous agent.
  • Reads under the definition's purpose. An agent whose definition names a purpose reads customer data under that purpose only. The instance owner must hold it.
  • Built-in agents. Six ready-made agents you set up with a few parameters: Anomaly explainer explains a metric alert, Campaign evaluator reports on a campaign after it is sent, Weekly campaign digest summarises the week's campaigns, Data quality watchdog flags drops in incoming data, Lead qualifier classifies web form submissions, and Privacy digest sends your data protection officer a weekly consent and erasure summary. Every figure in their emails is checked against the data they read.

Agents need the Agents licence package with the Custom agents or Built-in agents feature (or both), on top of AI assistants. Two new roles go with them: agents_editor builds definitions and runs instances, and agents_auditor reads instances, runs and the log. See Licences and roles.

AI models now takes an input and output price per million tokens and a currency for a text model, so agent budgets can be set in money. See AI models.

Privacy

  • Consent statistics. New counts of consent grants and withdrawals per purpose, per day or per week, without any profile ids. They are in the API and the AI assistant tools, and the Privacy digest agent emails them weekly.
  • Purpose names without dots. A new purpose name can no longer contain a dot. Existing purposes with a dot keep working, and filters, segments and reports on them now check the right consent.
  • Privacy mode off now behaves like audit mode: every value is returned, and reads are still checked and recorded in the audit log.
  • Members of a suspended tenant now get a clear "tenant suspended" refusal.

Schedules

  • Campaign, journey and audience-sync schedules now follow the same rule as connectors and agents. When a schedule sets both a day of the month and a day of the week, a day that matches either one runs it. Before, both had to match. The console explains this under the schedule.
  • A schedule that can never run, such as 30 February, is refused when you save it.
  • After missed runs, for example during maintenance, a schedule runs once for the latest missed time instead of once for each.
  • When clocks go back, a time that occurs twice runs once. When clocks go forward, a time that is skipped does not run that day.

See Schedules (cron).

Campaigns

  • Deleting an active campaign is refused, and so is deleting a template that a draft, active or paused campaign uses. The message names the campaigns.
  • Saving a journey checks the profile attributes, catalogue names and webhook conditions it uses. Test this request refuses a success condition that does not work.
  • A campaign stopped by one of its stop conditions shows in the change history as stopped by the system.
  • Performance insights and encrypted PDF attachments work again.
  • An activation destination keeps its settings when you edit it, and changing its ad account asks for the credentials again.
  • Campaign results now show a rate for each step of the funnel.
  • Campaign touches are still recorded when the tenant has reached its daily event limit.

Analytics

  • Metric alert emails are no longer lost when the metric's creator has no stored email.
  • The analytics_viewer role can now run and preview reports and metrics. It still cannot change them.
  • The metrics list can be searched by part of a name or description.
  • Signals and cohorts can no longer be renamed, and a cohort's entity type cannot change after it is created. Editing a cohort filter resets its member count.

Data and the API

  • Walk large audiences. New reads list every profile id, the profiles changed since your last call, the attributes and recent events of up to 100 profiles at once, and the events of one type in arrival order. Profile lists can also be paged with a cursor (after and next_after), which stays fast deep into a large audience.
  • Cached signal values. Profile and signal reads take signal_refresh=cached to answer with the stored values instead of waiting for stale ones to be recomputed.
  • Lookups by identifier and profile id together return the profile only if it has that identifier. Sending only half of an identifier is refused.
  • Dataset record upserts are all or nothing. If any record in a request is invalid, none of the records is written.
  • Events are listed newest first.
  • Clearer validation. Many saves that used to accept incomplete or contradictory settings are now refused with a 400 that names the problem: erasure, retention, identity rules, profile links, contracts, connectors, on-site settings and forms. Error messages keep their original capitals and read as sentences. If your integration matched on lower-case error text, compare without regard to case, or match on the status and data.reason. See Errors.
  • Unique names. Application names are unique within a tenant and content item names within a folder, ignoring case. A clash returns 409. Existing duplicates were renamed with their id in brackets.
  • Private addresses are refused. Providers, external systems and webhooks whose URL is a private or loopback address are refused when you save them, unless your platform operator has allowed that range or the tenant has a private network.
  • Browser calls from other sites to the API need your platform operator to allow the site's origin. The console is not affected.
  • Acting for a user. The user an application acts for must be given by email address.
  • Faster ingest and reads. Profile batches, signals, dataset records, event writes and event reads are much faster under load, and personalised on-site content answers faster.
  • API keys now see Campaign attachments according to their own personal-data level.

The API reference lists every endpoint.

Console

  • Shorter help. Hints and introductions are shorter in all eight languages. The longer explanation is behind the ⓘ icon next to a field.
  • Sessions. An expired session takes you to the sign-in page once and back to where you were. Signing in clears the previous user's acting purpose, searches and cached data. In strict privacy mode, a section that needs an acting purpose asks you to choose one instead of showing an empty page.
  • Unsaved changes. Cancel in a dialog with unsaved changes asks before closing, and the instance, account and segment editors warn before you leave with unsaved changes.
  • Profiles. Look up trims spaces and says which half of an identifier is missing.
  • Editing. Journey undo starts from the loaded journey. Uploading an asset whose name is taken gives it a numbered name. Campaign editors warn when an email subject contains a line break.
  • Permissions. Actions your role cannot perform, such as previewing a metric, are disabled with the reason. Home shows a welcome card with the areas you can reach when you cannot see the analytics overview.
  • Users. Edit user sends a password reset instead of setting a password.
  • Usage. A Right now strip shows current usage apart from the daily billing totals.
  • Small screens. Tabs, settings navigation and page headers no longer scroll sideways on a phone.
  • New logo. The console uses the Optima logo of optima-cdp.com.

AI assistants (MCP)

  • A new documentation section, AI assistants (MCP), explains how to connect Claude, ChatGPT, Claude Code and Cursor, which roles unlock which tools, how purposes and personal-data levels apply, and the full tool list with example prompts.
  • Signing in from an AI assistant and showing each key only the tools it may use work again.
  • New and extended tools: consent statistics, cursor paging and cached signals on profile queries, and a purpose on the signal history read.

SDKs

No SDK has a new release. These versions go with Optima v1.1.0:

On this page