Authentica
Book a demo

What's new

Every release, and what it changes

Authentica ships continuously. This is the running record of what reached production, written for the people who use it rather than the people who built it.

Latest release 2.11.13

2.11.13

Rehearse a workflow before it runs

A workflow can now be rehearsed against your live records without changing anything. The rehearsal walks each step and reports what would run, what would be skipped and why, which record each step would act on, and where the run would wait for approval, and Studio shows one for every workflow a proposed change touches.

  • Studio

    Workflow rehearsal

    A rehearsal is a dry run of a workflow against a proposed change and your organization's live records. It writes nothing and makes no model calls, and it reports step by step what would run, what would be skipped and why, which record each step would bind to, and where the run would stop for approval.

  • Studio

    Rehearsals in review and before Merge & apply

    Reviewing a change proposal in Studio shows a rehearsal of every workflow the change touches, and Merge & apply lists anything a rehearsal flags as a warning before you apply. Agents can rehearse a workflow too, so what a proposed change will do can be checked rather than taken on trust.

  • Agents

    Steps find the record their run just wrote

    A workflow step whose condition depends on a record written earlier in the same run checks that record, even when your organization holds several records of the same type.

2.11.11

Kate runs the work and brings back the answer

Kate now assigns work to the right specialist herself, relays its questions and results in your conversation, and lets you watch the specialist's steps as they happen. Every figure in an Analytics reply is checked against the results the platform holds, a draft code tool can be run once with a person's approval, and reviews are decided one piece of work at a time.

  • Agents

    Kate assigns the work and relays the answer

    Ask Kate for something and she assigns it to the right specialist, then relays its questions and results in the same conversation. Your answer to a question reaches the work that asked it, Kate acts within your own permissions, and the agent directory shows what each agent can reach and how many of its capabilities can run right now.

  • Agents

    Watch a specialist work, live

    When Kate hands work to a specialist, its steps stream into the conversation as they happen. A finished change proposal is posted to its conversation the moment it is ready, whether or not the chat is open.

  • Analytics

    Every figure in an Analytics reply is checked

    The platform checks each figure an Analytics reply states against the results it actually holds, from queries, the Vault, integrations, documents and code runs, at the precision the figure was written. A figure no result supports is named in a note under the reply.

  • Agents

    Run a draft code tool once, with approval

    A code tool still in draft can be run once from a Console conversation after a person approves that single run, and the outcome, completed, failed or declined, is reported back to the conversation. Kate can also run your organization's registered code tools, within each tool's own approval rules and your permissions.

  • Agents

    Reviews decided one piece of work at a time

    A task holds until every question it has put to a person is answered, and a review covering several pieces of work is decided on each one separately. A review can be requested from a conversation, and the decisions come back to it.

  • Analytics

    Analytics without a warehouse

    Catalog metrics, dashboards and reports are available to the Analytics agent in organizations without a connected warehouse, since only schema discovery needs one. When metrics exist but the Vault holds no records yet, the agent says every metric reads zero rather than building an empty report.

Also in this release

  • Plans, briefs, assigned work and decisions fold into pills above the composer once you move on, and decided items move to the side panel.
  • A group of calls where some failed reads as, for example, 4 succeeded, 1 failed, with red kept for when every call failed.
  • A running plan that needs you says so on its card, and Hide in the Waiting on you list is labelled as deciding nothing.
  • Every agent can read and write documents, and your AI forward-deployed engineer can write text documents.
  • Notifications can be closed, and a burst of them collapses into a single notifications button.
  • A large dashboard tile or workflow step result loads in full as soon as it is ready.
  • The user guide is reorganized, the Console has its own pages, and the Help menu and the Console's documentation link open the right place.
  • Security improvements.
2.11.0

Agents that know what they can reach

Every agent that carries out work can research the web by default, and each agent knows what it can reach in a conversation and why not, so it can tell you what it is not granted and where to go. Agents answer from the platform's record of the conversation, plans name the inputs each step needs, and Waiting on you counts the work that is waiting on a person.

  • Agents

    Web research for every agent that acts

    Agents that carry out work, including your organization's own agents, can search the web and read pages by default. Supervising agents, which direct work but never carry it out, do not get it, and it can still be turned off for any individual agent.

  • Agents

    Agents say what they can reach, and why not

    Each agent is told what it can reach in the conversation it is working in, connected integrations, the warehouse, datasets, documents and the web, along with the reason for anything it cannot reach and the route to it. When something is out of its reach, the agent says it is not granted that here and points to the agent or connection that can help.

  • Agents

    Answers from the record, not from memory

    Kate and the specialists she routes to are given what the conversation has caused, its plans, runs, escalations and proposals, from the same record the side panel shows. A reply that mentions a plan, run or escalation that is not on record is marked as such, and a task result posted to the chat carries the name of the agent that did the work.

  • Agents

    Plans name their inputs

    Each step of a plan shows the inputs it needs by name. A plan missing an input, or passing one a step does not accept, is refused with the input named, and a step given to an agent that cannot take it is refused with the agents that can.

  • Platform

    Waiting on you means waiting on you

    Waiting on you counts the work held by a person: a blocked task, a pending review, a manual start or a failure. Resuming a task that has already finished tells you so, and Acknowledge & resume on a review with attached work asks before it goes ahead.

Also in this release

  • Platform-managed agents are labelled Managed by Authentica, available to every team, and pick up assigned tasks from every team they serve.
  • Every task an agent receives carries its procedure and the part of your operating model it needs, whether it came from a template, a chain or a workflow.
  • The collapsed sidebar shows your team's initials and keeps its height through a collapse.
2.10.9

Authentica on your phone

The product now works on a phone and a tablet. Navigation opens as a slide-out panel, the Console runs full screen with its work list a tap away, and dialogs, side panels and tables fit the screen they are on.

  • Platform

    Use the Console from a phone

    On a phone the Console opens full screen from its pill, and the list of your work opens as its own panel from the top bar and closes when you pick an item. Send stays on screen however long the model name is, and the thread stays pinned to the newest message while a reply streams in.

  • Platform

    Navigation built for a small screen

    Below tablet width the sidebar opens from a button as a slide-out panel and closes when you navigate. Breadcrumbs stay on one line, search collapses to a button that opens the field across the header, and a tablet rotated back to landscape returns to the sidebar layout it had.

  • Platform

    Everything fits the screen it is on

    Dialogs and side panels size to the screen and scroll inside, with a close button and the page dimmed behind them. Buttons and menu rows are larger under a finger, tapping a text field keeps the page at its size, and floating buttons clear the home indicator.

Also in this release

  • Long record names wrap to two lines with the full name on hover, and tab strips and filters scroll sideways instead of the whole page.
  • Notifications read as one line per item, grouped by day.
  • Studio is built for a larger screen and says so on a phone, with the option to open it anyway.
  • In the collapsed sidebar, a Vault organized into business domains shows one button per domain and unfolds the same sections as the wide sidebar.
2.10.0

Analytics that researches before it answers

Every analytics question now runs under a research protocol: interpret, discover sources, inspect the schema, plan, explore, analyze, answer. A run can be watched beside the chat, a dashboard can be built from a conversation, and the Conversation Ledger shows everything a chat caused.

  • Analytics

    The Analytics agent as a research layer

    Each analytics turn follows a platform-authored protocol: interpret the question, discover sources, inspect the schema, plan, explore, analyze, answer. A question is framed against your organization's catalog before anything runs, and a request that names nothing measurable gets one clarification rather than an invented query. The agent is briefed on every data surface the organization has, warehouse, connected integrations, the Vault and uploaded documents, so a missing warehouse leads to the next surface rather than to “not available”.

  • Analytics

    An empty result is a diagnosis, not a finding

    An empty or all-zero result carries a diagnosis, either the source has no data or the filters excluded everything, and the identical query is refused on a repeat. The conversation keeps a ledger of what was tried, a reply resting on one unexamined empty result is either pivoted or labeled unverified, and warehouse results link back to the conversation that proposed them so a report can cite them.

  • Platform

    Watch a run beside the chat

    A Run tab in the side panel shows status, progress, tasks and live tool waits without leaving the conversation. Hovering a record key, run, task or deliverable link in a thread opens a summary card beside it.

  • Analytics

    Build a dashboard from a conversation

    Ask for a dashboard in the Console and the finished dashboard lands in the Dashboards tab, linked from the chat and from the run's details.

  • Platform

    The Conversation Ledger

    One view of everything a chat caused, decisions, plans, runs, documents, reports and pull requests, behind the Waiting-on-you block and the side panel.

Also in this release

  • Documents and other deliverables an agent creates arrive as usable links or attachments.
  • An approval answered inline in the Console hands its result back to the conversation on a fresh turn, and a placed phone call reports its outcome and transcript to the chat when it ends.
  • Studio re-anchors a positional change that no longer matches its element instead of failing, and rebases a stale personal draft onto the shared working copy on load.
  • Superseded plan notices can be dismissed and stay dismissed.
2.9.0

Deleting is an admin action

Editors create, read and update; only an Admin can delete. Sign-off requests land with the people who can act on them, a team document can be filed into Google Drive by reference, and software change requests can be governed as a lifecycle in the operating model.

  • Security

    Editors cannot delete

    Platform-wide, an Editor creates, reads and updates. Deleting a case or a record is an Admin action, so a team can work its cases freely without any member being able to remove one.

  • Agents

    Sign-off requests reach the people who can act

    A specialist that needs an approval in the course of a conversation raises it, and a team-level sign-off lands in Needs Attention for every member of that team. A pending-approval state is never shown with nothing to approve.

  • Integrations

    File a team document into Google Drive

    An existing team document can be filed into Google Drive by reference through a declared file operation, with no re-upload.

  • Platform

    Software change requests as a governed lifecycle

    A shared software-delivery module governs change requests as part of the operating model: draft-to-merge states, governed transitions, phase task types, role blueprints and event-chained workflows, available to any dedicated environment.

Also in this release

  • A lifecycle action whose transition is not open is refused with a typed reason and writes nothing.
  • When the Ontology Author changes an element of a shared module, it produces a configuration patch rather than a copy, so your organization stays on the shared version.
2.8.0

Web search for agents, and governed software delivery

Agents can search the web and read pages when your organization grants it. Engineering teams get a governed software-delivery workflow with work-unit states, review verdicts and repair flows, and the Ontology Author repairs a rejected proposal before it reaches you.

  • Agents

    Web search and page reading, by grant

    Platform agents can search the web and fetch pages. Worker agents get the same only when a runtime grant says so, so an agent's reach onto the open web is a decision your organization makes, not a default.

  • Platform

    Governed software delivery for engineering teams

    A software request routes to an engineering workflow with work-unit states, review verdicts, repair flows and event-triggered execution. The platform supplies the available code projects and inputs rather than asking the requester for them.

  • Studio

    The Ontology Author repairs before it proposes

    Authoring validates the exact location of every change against the current operating model and repairs a rejected proposal before presenting it for review, so what reaches you is a candidate that applies.

Also in this release

  • Emails from a dedicated deployment carry that deployment's name, logo, theme and support address.
  • Transient failures in a run are retried automatically within a per-task budget rather than escalated to a person.
  • Food safety and quality workflows use configurable terminology, grouped by domain in the Vault.
2.7.0

Agents that read the guide, and a Stop that stops

Every agent can answer “how do I…” from the product documentation, section by section. Stop halts a whole run in one action, Studio's history is a single timeline, and the Console is easier to find your way around.

  • Agents

    Every agent can read the user guide

    The documentation is available to every agent as a retrieval service: outline, ranked search, one section at a time. A “how do I…” question is answered from the docs rather than routed away or answered from memory.

  • Agents

    Stop actually stops

    Stop works on every task that has not finished, and a whole workflow run can be stopped in one action, with its pending approvals and escalations cleaned up. The run reads as cancelled.

  • Studio

    Studio history as one timeline

    The History tab interleaves applies, rollbacks and commits newest-first by when they happened. Accept shows that it is working, and Merge & apply is a single dialog from the first click until the run starts.

  • Platform

    Console navigation

    Browser Back and Forward walk the rows you opened, every unsettled run has Stop, escalation review opens in the side panel, @-mentions survive spaces in a name, and live status moves while a run is running.

2.6.0

A real path forward for blocked work

When a task blocks, the escalation offers structured options: retry now, retry with changes, relax a constraint, approve and resume, each recorded in the audit trail. Organizations can also let trusted agents run read-only analytics queries unattended, off by default.

  • Agents

    Structured escalations

    A blocked task's escalation presents actionable options rather than a free-text note: Retry now, Retry with changes to update parameters first, constraint toggles to relax a rule, and Approve & Resume. Every decision is auditable.

  • Security

    Approvals that match the plan, or nothing runs

    An approval card shows the effective parameters the work will run with. If what is about to execute differs from what was approved, execution is refused. Chat approvals reach the blocked task directly, with a link when the approval has to be given elsewhere.

  • Analytics

    Unattended analytics under a trust floor

    An organization can set a minimum trust level at or above which agents run read-only analytics queries without a person approving each statement; one click per query made large data reviews impractical. It is opt-in per organization and off by default. A policy approval is recorded distinctly from a human one in the audit trail, the read-only limits and catalog allowlists still bound every query, and a platform kill switch restores human approval everywhere.

2.5.1

Approve from the chat, several at once

When work you dispatched pauses for an approval, the exact approval card appears in the conversation with Review, Approve and Deny in place. Several approvals from one chat can be decided in a single action, and Studio says where an accepted change went.

  • Agents

    The approval card comes to the conversation

    When work dispatched from a chat stops for an approval, the chat shows the same card the task page shows, with the query, the tables, the estimate and the plan, and Review, Approve and Deny in place. A pill beside the composer counts what is waiting and jumps to it.

  • Agents

    Approve N selected

    Two or more approvals raised from one chat can be approved together. Each plan is still approved exactly as written; only the submission is batched.

  • Studio

    Studio tells you where an accepted change went

    After an accept, the panel says whether the change landed in the shared working copy or in a single draft, a resolved review renders as history rather than pending, and an unknown or downgraded version is named up front in the compose message.

2.5.0

The Console is the front door

Open Authentica and you land in a fresh conversation, ready to start work. One toggle moves between Console and Delegate, notifications count only what needs your action, and a chat that dispatched work tells you the moment that work is waiting on someone.

  • Platform

    Start from a blank conversation

    The Console opens on a fresh, unsaved conversation instead of whichever task last moved. One Console | Delegate toggle, one Notifications surface and a Settings shell, and every chat shows the work it created.

  • Platform

    Notifications count what needs you

    The badge counts only items that need your action. Unread but informational items show as a dot, so a number on the badge always means something is waiting on you.

  • Agents

    Know when dispatched work is stuck

    When a task started from a chat is blocked or waiting on a person, its run chip takes the attention tone, hovering shows the task's own reason with a link to it, and the sidebar marks the conversation as waiting on you. Force retry reopens the chat that started the task.

Also in this release

  • An agent's question always offers at least two real choices, enforced by the platform whatever the model does.
  • Routed specialist turns record which model answered.
2.4.0

From an uploaded spreadsheet to records, without the detour

Upload a document and turn its rows into records directly; the file never has to pass through the model. Agents read large spreadsheets and PDFs page by page, a platform-managed Code Author joins the roster, and the Console shows live activity in a side panel.

  • Agents

    Bulk ingest from an uploaded document

    Upload a spreadsheet or a PDF and ask for its rows to become records. The platform reads the file and writes the records directly; the contents never pass through the model, so nothing is paraphrased or dropped on the way in.

  • Agents

    Meet the Code Author

    A platform-managed Code Author handles code work for your organization, alongside Kate, the Analytics agent, the Ontology Author and your AI forward-deployed engineer.

  • Platform

    Live activity beside the conversation

    An editor-style side panel shows what an agent is doing while it works, and a resumed turn continues in place rather than as a new message.

Also in this release

  • Agents read large spreadsheets and PDFs page by page, and every agent has the standard document tools by default.
  • Studio proposals are built only from changes that have passed validation.
  • Validation reports every missing field in one pass, and browsing the operating model shows descriptive entity details rather than bare keys.
2.3.0

A governed code workspace for every organization

Every organization has its own code repository, and the Analytics agent can write and run read-only code against the datasets that organization has been granted, and nothing more. Access to data fails closed.

  • Analytics

    The Analytics agent can write and run read-only code

    For a question no pre-built query answers, the Analytics agent can write code, run it read-only against your data in an isolated sandbox, and bring the result back into the conversation.

  • Security

    One code repository per organization, access failing closed

    Every organization is provisioned with its own code repository, so code an agent writes for you lives with your organization. Dataset access is granted per organization and fails closed: a dataset that has not been granted is not reachable, whatever the code asks for.

Also in this release

  • Analytics query failures surface feedback in the conversation rather than failing silently.
2.2.3

Merge and apply from Studio

A reviewed change to your organization's operating model can be merged and applied from Studio in one action, with no release cycle and no waiting on us. Platform-managed agents serve every team in the organization, and the Analytics agent builds dashboards and reports on request.

  • Studio

    Merge & apply in one action

    When a change to your organization's operating model has been reviewed, one action merges it and applies the merged version to your organization. It is scoped to your organization's own content, anything outside it is refused before the merge, and it is enabled per environment.

  • Agents

    Platform-managed agents serve the whole organization

    Kate, the Analytics agent and the other platform-managed agents serve every team in an organization, with reads and integration calls scoped to the team of the conversation they are working in.

  • Analytics

    Dashboards and reports from the Analytics agent

    The Analytics agent handles ad hoc queries, builds dashboards and reports, and describes the structure of a synced table in readable terms.

Also in this release

  • An apply that fails validation returns the specific finding, and a breaking change is reported as such so a deliberate force-apply can be offered.
2.2.2

Ask the Analytics agent anything about your connected data

The Analytics agent reads the data sources your organization has connected and answers ad hoc questions with themed reports and dashboards. An applied operating model can be rolled back to an earlier version with a preview and full lineage, and Studio sessions carry across applies, reviews and rollbacks.

  • Analytics

    Ad hoc questions against live sources

    The Analytics agent is no longer limited to what was already loaded. It reads the organization's connected data sources and answers an ad hoc question with a themed report or dashboard.

  • Studio

    Roll back an applied operating model

    Rolling an applied operating model back to an earlier version is a first-class operation: preview it, apply it, and keep the lineage. Studio sessions carry across applies, pull requests and rollbacks, a session whose review has merged or closed retires into a fresh successor with its draft carried over, and large operating models open in milliseconds.

  • Agents

    Address your AI forward-deployed engineer from inside a thread

    Mentioning the AI forward-deployed engineer inside an existing conversation is treated as an address, so a build request routes correctly from within a thread and when assigning follow-up items.

Also in this release

  • Agents read large spreadsheets and Word documents directly, truncating an oversized range instead of abandoning the read.
  • A person to whom a console review has been escalated can confirm it themselves.
2.1.7

Shared connectors for the whole organization

An integration your organization depends on can be connected once, at team level, so it is never tied to one person's login. One person connects it under Team / Shared and every governed workflow uses it, and an action that points at an integration operation that does not exist is refused when it is authored rather than when an agent runs it.

  • Integrations

    Team-level integrations

    A team-level connector appears under Team / Shared, and one person with integration-manage permission connects it for everyone. An organization-wide sync runs under the shared team identity and keeps running if the person who set it up leaves or revokes their access. HubSpot joins NetSuite as a team-level integration, and which connector types are team-level can be changed without a deploy.

  • Studio

    Bad integration targets are caught at authoring time

    An action pointed at an integration operation that does not exist is refused when it is authored, with the reason, rather than when an agent runs it. The check runs only against targets an author adds or changes, so existing content keeps applying.

Also in this release

  • Custom connectors built through the custom-integration wizard connect cleanly.
2.1.5

Agents that hand you files, and act through your own connections

Ask for a spreadsheet and get a downloadable file in the conversation. The console assistant can read and, with your sign-off, write through the integrations you have connected personally, and every approval it needs comes to you and reports back. Each organization decides who may apply changes to its operating model.

  • Agents

    Real files, delivered in chat

    Ask the console assistant for a CSV or an export and a downloadable file card appears in the conversation. Created files share the same lifecycle as uploads, so they can be filed into the Vault or against a record through the existing placement review.

  • Integrations

    The personal connector rail

    The console assistant can read and write through your own connected integrations: any connector, the exact operation named per call. Every write pauses for an approval that shows the precise payload before it goes out under your name.

  • Agents

    Approvals reach the person who asked, and report back

    An approval raised on your behalf is assigned to you and appears as a card in your chat and your inbox. Once decided, the agent is told the real outcome, executed or failed and why, instead of going quiet after “pending”.

  • Security

    Per-organization control over who may apply changes

    Each organization decides who can apply changes to its operating model, and whether authors may accept their own Studio proposals. Self-acceptance is allowed by default in dedicated environments and off in the shared environment; the organization's own setting wins either way.

Also in this release

  • Every person gets a private personal team that is not visible to the organization.
  • Integration errors are readable: a missing credential raises a proper escalation, and provider errors surface as plain messages.
  • Analytics: an invalid metric or dimension definition returns an actionable validation error.
2.1.4

Review together in Studio

A Studio proposal is pinned to the operating model it was written against, so every reviewer sees the same change, and accepting it lands in the organization's shared working copy. Agents can also draw on what earlier runs learned, and documents attached to a record can be downloaded from it.

  • Studio

    Every reviewer sees the same diff, and acceptance lands in the shared copy

    A proposal pins the version of the operating model it was written against, so reviewers no longer see a diff relative to their own editor. Accepting writes into the organization's shared working copy rather than the accepter's private draft. Accepting against a base that has since moved, or accepting your own proposal, is refused with an explanation and stays retryable, and every proposal shows who proposed and who accepted.

  • Agents

    Agents can answer “what did we learn from this?”

    Lessons recorded against a task or a workflow run are readable by an agent, and the task view reads the same source, so the panel and the agent cannot disagree about the same work.

  • Platform

    Download documents attached to a record

    Any document attached to an entity can be downloaded from that record's Documents tab.

Also in this release

  • Studio's review lifecycle is steadier from proposal to apply: resolved proposals stay resolved and Apply is available whenever it should be.
2.1.3

Confirm the build before it starts

Your AI forward-deployed engineer now shows you a build brief and waits for your go-ahead before the Ontology Author starts work. Agents can produce finished documents, the Vault is organized by business domain, and a problem you raise in a conversation becomes a tracked report rather than a message that scrolls away.

  • Studio

    A reviewable brief before any change is built

    Ask for a change to your operating model and you get a build brief to approve first. Once you approve, the Ontology Author's progress streams back into the conversation, and the resulting proposal separates what you asked for from drift the organization had already accumulated, so an additive request never reads as though it will delete existing work.

  • Agents

    Agents can produce documents

    Ask an agent for a summary, a report or a brief and it can produce a finished document as a PDF, Markdown or HTML, rather than pasting text into the chat.

  • Platform

    Report a problem without leaving the conversation

    When your AI forward-deployed engineer cannot resolve something you have hit, it files the report with our team, tells you it has been filed and assigned, and groups repeat reports of the same problem rather than duplicating them.

  • Platform

    The Vault is organized by business domain

    The Vault sidebar groups documents by business domain, scoped to the teams you belong to, and explains an empty list instead of leaving it ambiguous.

Also in this release

  • One consistent answer to “which actions can I use?” on every surface. A denied action shows why: the gate, the agent's trust level, the supervisor ceiling, or a permission you do not hold.
  • Conversations stay with the agent you are talking to: answering an agent's question keeps you with that agent, and every turn carries its routing history.
2.1.2

Meet Kate, your orchestrator

The agent that coordinates your specialists gets a name and a plainer voice, capability lists track the systems you have actually connected, and change protection travels with every edit.

  • Agents

    The orchestrator presents as Kate

    The agent that coordinates work across your specialists has a name and a consistent voice, so it is clear who you are talking to when a request is being routed rather than executed.

  • Platform

    Self-serve review, in plain language

    Review copy across the console is written for the person doing the reviewing. Handoffs to our engineering team route with an explanation of what happens next.

  • Integrations

    Capabilities match your connected systems

    An agent's capability surfaces are drawn from the platform actions your workspace has connected, so what an agent offers is what it can execute.

Also in this release

  • Version guarantees hold across every authoring path, Studio or command line.
2.1.0

Bring your own files into the conversation

Attachments travel with the work. Upload a document in the console and the agent you dispatch can read it, with access scoped to that agent and that task.

  • Agents

    Uploaded files reach the agent doing the work

    Drop a spreadsheet, contract, or report into a conversation and it is delivered to whichever agent picks the work up. Read access is granted for that dispatch alone, scoped to that agent and that task.

  • Studio

    Address the Ontology Author directly

    Mention the Ontology Author in a conversation and the request routes straight to the managed authoring agent, so operating-model changes start with the specialist that owns them.

Also in this release

  • A single consistent error format across the agent-facing API, so integrations can handle failures uniformly.
2.0.0

The governed agent console

Authentica 2.0 turns the agent surface into a place operators actually work: ask for something in plain language, watch the agent do it under your governance rules, and approve the parts that need a human. Every capability an agent holds is declared, visible, and enforced at run time.

  • Agents

    Ask for work in plain language

    The console routes a request to the right specialist without you naming one. Type what you need, and the platform picks the agent, confirms the interpretation when it matters, and hands the work over. A `/` picker is there when you would rather choose directly.

  • Agents

    Conversations that survive the tab closing

    Agent responses stream durably. Close the browser mid-run and the work carries on; reopen and the thread is where you left it. Completed results, reasoning, and outcomes flow back into the same conversation.

  • Security

    Every capability is declared, and the ceiling holds

    A capabilities panel shows exactly what each agent may do at each trust level, derived from its autonomy gates rather than described by hand. Supervisory agents can never execute change or communication actions — an absolute limit with no override. Human-initiated turns carry that human's permissions.

  • Agents

    Approve and resume in the flow of work

    Approval requests appear live as they are raised, and approving an action executes it there and then. Review surfaces distinguish a supervisor's sign-off from a human's, and escalations can be answered inline so the work resumes from where it paused.

  • Analytics

    Governed analytics on your own warehouse

    Agents can discover warehouse schema through an allowlisted, governed path, build dashboards and reports from a natural-language brief, and produce charts inline in the conversation. Query approval is available per invocation for teams that want a human in front of every read.

  • Platform

    Agents that write and run reviewed code

    For work no pre-built action covers, agents can propose code, run it through continuous integration, and surface the verdict — including failure detail — for review before anything is promoted. Code runs read-only against explicitly declared data, with no network egress.

  • Studio

    Studio for authoring the rules

    Author your operating model with live validation from the same engine that enforces it at run time, see the workflow your agents will actually follow, and ship every change through review.

Also in this release

  • Freshdesk actions execute through the integration proxy.
  • Outlook send and reply-all thread context, with attachments carried into workflow steps.
  • Scoped access profiles, per-environment control over self-service registration, and hardened enterprise invitation onboarding.
  • Inline document viewer for chat attachments.
  • Dark-mode support for agent-generated charts.
  • Faster delegate dashboards and task pages on large workspaces.