Sign in

News & Blog

Research, engineering, and ideas from Actava.ai.

Videos

117 videos

VideosWatch video

ACTAVA | CURA | Healthcare Workflows in Practice

CURA executing inside the EHR without breaking anything. Frontier models are generalists. Healthcare punishes generalists. The work is conversational one minute and transactional the next, and the cost of a confident wrong answer is measured in patient harm, denied claims, and audit findings. That is why Actava.ai built CURA, a 1-trillion-parameter model trained for clinical and administrative healthcare work, and why we tested it where it matters. Across six of the hardest healthcare benchmarks, CURA outperformed frontier models. This example focuses on where most models quietly fail, and where the long tail of healthcare administration actually lives. MedAgentBench v2 requires retrieving patient data through structured FHIR queries, documenting vitals and clinical notes, and placing medication, lab, and referral orders through correctly structured write operations. Tasks average two to three chained actions, mirroring routine inpatient and outpatient work, and a single malformed JSON payload or invalid API call counts as a failure. There is no partial credit for good intentions. CURA's performance here means agents that reason not only about the record, but work the record safely and in the required format. That is the capability every administrative workflow beyond prior authorization depends on, and it is the reason Actava.ai focuses on the complex, document-heavy, cross-system work that generic AI cannot reliably execute: the workflows that are hard, policy dense, long running, and unique to your organization. One model now covers the full arc of healthcare work, from an uncertain patient conversation through specialist reasoning to a correctly formatted order in the EHR. Deployed through KORA, governed by CHRYSO, and measured continuously with χ-BENCH, CURA gives healthcare and life sciences teams a workforce-grade model for the thousands of workflows the industry has never been able to automate.

Watch ↗

VideosWatch video

ACTAVA | CURA | Clinical Reasoning in Practice

CURA delivering specialist-grade reasoning, not trivia recall. Frontier models are generalists. Healthcare punishes generalists. The cost of a confident wrong answer is measured in patient harm, denied claims, and audit findings. That is why Actava.ai built CURA, a 1-trillion-parameter model trained for clinical and administrative healthcare work, and why we tested it where it matters. Across six of the hardest healthcare benchmarks, CURA outperformed frontier models. This example focuses on the reasoning that utilization review, clinical documentation integrity, prior authorization criteria matching, and medical affairs work all depend on. MedXpertQA Text spans board-style questions across 17 specialties and 11 body systems. The cases read like real vignettes, with history, labs, and family history feeding a diagnostic and treatment decision that requires multi-step inference. MedXpertQA Multimodal goes further, pairing a medical image from radiology, pathology, or dermatology with patient records and exam results. The image alone is never enough. The model has to integrate visual findings with clinical context across multiple stages of the diagnostic process, which is exactly how imaging-adjacent workflows behave in production, from radiology worklist support to dermatology e-consults to pathology second reads. AgentClinic is the agentic test, and it is where reasoning has to hold up under pressure. A doctor agent starts with almost nothing, interviews a simulated patient, decides which tests to order, interprets the results, and commits to a diagnosis within a limited number of turns. Some cases add medical imagery. Others inject cognitive-bias perturbations such as anchoring or patient-introduced bias, designed specifically to degrade accuracy. CURA held up. For anyone building autonomous clinical agents on KORA, that is the difference between a demo and a deployable Non-Human Resource that manages its own line of questioning. Deployed through KORA, governed by CHRYSO, and measured continuously with χ-BENCH, CURA gives healthcare and life sciences teams a workforce-grade model for the thousands of workflows the industry has never been able to automate.

Watch ↗

VideosWatch video

ACTAVA | CURA | Patient Care in Practice

CURA talking with patients the way clinicians actually do. Frontier models are generalists. Healthcare punishes generalists. The work is conversational one minute and transactional the next, and the cost of a confident wrong answer is measured in patient harm, denied claims, and audit findings. That is why Actava.ai built CURA, a 1-trillion-parameter model trained for clinical and administrative healthcare work, and why we tested it where it matters. Across six of the hardest healthcare benchmarks, CURA outperformed frontier models. This example focuses on the messy front door of healthcare. HealthBench Hard tests exactly that: a patient describes vague or conflicting symptoms over a back-and-forth chat, and the model must infer the missing context before it can answer. It has to spot red flags and escalate to urgent care. It has to hedge when the evidence is thin rather than overcommit. And it has to perform in global health and resource-limited settings, not only in a well-resourced US academic center. For virtual care teams, nurse triage lines, and member services, this is the whole job. CURA's performance here means patient-facing agents that ask before they answer, escalate when they should, and stay honest about uncertainty. HealthBench Professional narrows further to the non-negotiables, testing behaviors physicians broadly agree are mandatory or forbidden, such as correctly recommending emergency referral when warranted, with near-zero error tolerance. That is the benchmark that maps directly to governance. When a health system asks how it can know the agent won't miss the thing it must never miss, CURA's results are the answer, and CHRYSO turns that answer into an auditable control. Deployed through KORA, governed by CHRYSO, and measured continuously with χ-BENCH, CURA gives healthcare and life sciences teams a workforce-grade model for the thousands of workflows the industry has never been able to automate.

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | Agents Built by Citizen Developers

Non-technical people co-build governed AI agents just by talking. "Shadow AI," the unauthorized use of consumer AI tools, is a governance risk. At Actava.ai, we also see it as a signal of internal demand. Citizen developers, meaning domain experts without coding backgrounds, are now the primary engine of progress. Forward-thinking companies are shifting from restricting unauthorized usage to empowering these builders within a governed, secure framework. As businesses integrate autonomous agents to handle high-volume tasks, the nature of work changes with them. The modern professional is evolving from a doer into an AI Agent Supervisor, where one leader orchestrates the productive output that previously required a team of five. This doesn't replace professionals. It augments their roles, freeing them to focus on oversight and strategy while agents handle execution. That forces a strategic choice: are you selling the tool, or are you selling the work? Selling software puts you in a race against large language models. Selling the work means owning the outcome, whether in data processing, documentation, or compliance. By rebuilding operating models around AI-native delivery, businesses become faster and more efficient. The people who understand healthcare workflows best should be able to shape the AI agents that support them. Actava.ai makes that possible by embedding industrial-grade agent design, evaluation, and governance into a platform business and operational teams can use directly. Guided natural-language building, built-in guardrails, evaluation rubrics, and expert deployment support let non-technical experts move from workflow insight to deployed agent without sacrificing safety, and without waiting on scarce technical resources.

Watch ↗

VideosWatch video

ACTAVA | Release | KORA v6 (June 2026)

Version 6 of the Actava.ai platform shipped in July 2026. Reach. Improvement. Proof — 58 items, behind one idea. In the release notes' own words: V6 turns KORA agents into a workforce that reaches members directly, improves itself under governance, and proves its value in dollars. 58 items across six categories. It starts with the model. CURA is a one-trillion-parameter model specialized for enterprise healthcare, with inference costing 20 to 100 times less than the frontier alternatives, full organizational ownership, and private deployment inside your own VPC. Then the channel. Agents send and receive SMS through Twilio or AWS End User Messaging, and an inbound reply resumes the agent's session rather than starting a new one. Opt-out compliance is built in, and a human approval can be routed over the same channel. And rather than every team rebuilding the same agents in silos, there is a library. Published listings with orchestration blueprints, a taxonomy by domain, global search, and cross-organization cloning of agents and their workspaces. Then the part that makes self-improvement safe. An agent proposes revisions to its own configuration, drawn from its run history. An operator reviews each one side by side against what is live. And an evaluation gate blocks anything that regresses before it can be promoted, so the loop can run unattended without the quality going quietly backward. Underneath that, versions became immutable. Test and prod badges are gone, replaced by draft and released versions that carry their own release notes, and existing agents were backfilled with snapshots, so the history starts full rather than empty. Awaiting-approval and paused became first-class run statuses. A paused run shows its state, appears in the live monitor, and resumes on approval — and the reaper treats it as paused rather than stuck, which is the difference between a queue and a leak. And then the proof. V6 captures real business events — staff time displaced, converted into dollars, with cost bindings and ratios — and aggregates them per period into KPIs that roll up into an organization-wide ROI portfolio. That is the groundwork the ROI engine builds on. Every user who opens KORA has questions the product itself should answer. So Ava is built in: one fixed identity that answers from the product, runs guided tours, builds agents and takes governed actions, with role-scoped tools, an organization kill-switch, and every system session recorded. And the models are callable from outside. The inference gateway exposes OpenAI-compatible endpoints behind key authentication, with a concurrency limiter and daily token counters enforcing per-key limits — a trial key stops at its dollar budget rather than running past it. Four things before you upgrade. SMS is off by default. Webhooks are opt-in. The default model moved across the chat tier, the wizard, and the evaluation judge. And versioning migrated, so any bookmark pointing at an old promote route needs updating. Reaching people. Improving itself. And proving what it was worth. Full release notes: https://actava.ai/news/actava-platform-release-notes-v6-july-2026

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | Google Slides

Your KORA agent builds board-ready decks in Google Slides while you stay in the conversation. The Google Slide Tool lets a KORA agent create, populate, and update presentations directly in Google Slides, pulling from the same clinical, operational, and financial context the agent already works with. Instead of exporting numbers into a spreadsheet and hand-building slides the night before a committee meeting, you ask the agent, and the deck assembles itself in your Google Workspace. The core insight: healthcare teams don't have a data problem, they have a translation problem. The information for a quality review or a payer QBR already exists. It just lives in the agent's reach rather than in a formatted, shareable story. This tool closes that last mile, turning a working conversation into a living presentation your colleagues can open, edit, and present without ever leaving the tools they trust. It starts the way everything in KORA starts, with a plain-language ask. Build a slide summary of last month's readmission trends by service line. The agent gathers the figures, cohorts, and comparisons it already has access to and shows you its plan before it builds, so the deck is grounded in real data rather than a blank template you have to fill. A new presentation appears in Google Slides with a title slide, structured sections, and charts, formatted rather than raw. Refinement stays conversational. Add a slide on the cardiology outlier. Make the trend chart quarterly. The deck updates in place, with no reformatting by hand. And because the result is a standard Google Slides file, it is shareable, commentable, and editable by colleagues with the right permissions, which means it fits your existing governance, review cycles, and collaboration norms rather than working around them. Next cycle, it is one line. Refresh the readmissions deck with March data. The recurring report that used to eat an afternoon now keeps itself current. The deck isn't the deliverable. The decision is.

Watch ↗

VideosWatch video

ACTAVA | χ-BENCH | Benchmarking Agentic AI for Healthcare Administration

χ-BENCH is the first benchmark to test agentic AI on healthcare administration as it actually runs. χ-BENCH evaluates 30 models and agent-harness combinations across provider prior authorization, payer utilization management, and RN care management. Its tasks are grounded in a 1,290-document managed-care operations handbook, making it one of the largest and most realistic evaluations of agentic AI in a policy-dense enterprise setting. Unlike benchmarks that test narrow capabilities such as clinical Q&A, FHIR calls, GUI clicks, or customer service interactions, χ-BENCH measures whether agents can complete full healthcare transactions that involve dense policy interpretation, multi-role coordination, hidden state, and provider-payer handoffs. The results expose a stark reliability gap. The best agent setup solved only 28% of complex tasks on the first try. No agent cleared 8% under consistency testing. On nearly a third of tasks, every model-agent combination failed. In a realistic provider-payer workflow, performance fell to 0% at the handoff. That evidence is backed by deep clinical and academic expertise. The review was anchored by T.Y. Alvin Liu, MD, of Johns Hopkins Medicine, and Hank Capps, MD, of Wellstar Health System, with authors from Stanford, CMU, UCSD, Yale, Salesforce AI Research, UW, Oxford, Brown, Emory, USC, and other leading institutions. Why it matters: U.S. healthcare spends about $1 trillion a year on administration, including roughly $200 billion on claims, payments, patient collections, and prior authorization. χ-BENCH makes the agentic AI reliability gap measurable. Even frontier agents still miss most workflows, fail consistency tests, and break down during real handoffs. Healthcare needs AI infrastructure that performs consistently, deterministically, and safely across policy-dense, end-to-end workflows. The defining question for healthcare is not whether agentic AI will work. It is whether the industry can reliably detect when it does not.

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | Agent Login As Viewer

See KORA exactly as your users see it, before they do. Agent governance only works if you can verify it. You can define user groups, scope agent access, and set role-based permissions across an organization, but the question that matters to a compliance officer is simpler than any config screen: what can this particular person actually reach? Emulated user views answer that by letting an administrator view the platform as a specific user or role sees it, without borrowing credentials and without guessing. That matters most at the moment permissions change. A new care management team comes onboard. A consultant needs access to one organization but not another. An agent moves from pilot to production and its access group is rescoped. In each case the risk is the same: a permission that looks right in the admin interface but behaves differently in the user's hands. Emulating the view turns that from an assumption into something you can see and screenshot. It also makes support and enablement faster. When a citizen developer reports that an agent isn't appearing for their team, an admin can reproduce the exact view rather than trading screenshots back and forth. When you are rolling out a new agent to a large group, you can confirm the experience each role will get before anyone opens a ticket. The same view that helps you troubleshoot is the one that helps you train. This sits alongside the rest of KORA's governed deployment model: group-based agent access control managed from one admin interface, multi-organization membership with separate roles and data scopes per environment, agent versioning with governed promotion, and human-in-the-loop approval checkpoints on high-stakes actions. Emulated user views are the verification layer over all of it. Governance you cannot inspect is a policy document. Governance you can see from your user's seat is a control. ——— Learn more Actava.ai: https://actava.ai KORA — agentic development lifecycle suite: https://actava.ai/why-kora CHRYSO — agentic governance & compliance suite: https://actava.ai/chryso χ-BENCH — AI simulation & benchmarking suite: https://actava.ai/benchmarks CURA — 1T-parameter healthcare model: https://actava.ai/cura From production workflows to customer-controlled intelligence. #Actava.ai #HealthcareAI #AgenticAI #AIOrchestration #AISafety #AIGovernance #AgentLifecycle #AIPilotToProduction #AgenticFuture #HealthTech

Watch ↗

VideosWatch video

ACTAVA | Agentic Workforce | Agents Built with Cost Controls

Turn every agent run into a graded, explainable, fixable result, with no data science degree required. Building an agent is easy. Trusting it is the hard part, especially in healthcare, where a plausible-sounding wrong answer is a real risk. Evaluate, Diagnose, and Optimize close that gap by turning each run into a graded, explainable, and improvable artifact instead of a black box. Evaluate scores every run against six built-in rubrics: Task Completion, Tool Correctness, Faithfulness, Answer Relevancy, Role Adherence, and Safety. Each score carries a plain-language definition and a pass or attention indicator, and you can open any rubric to see how the score was attributed across steps and tool calls. The number stops being a mystery. You see exactly which step dragged Faithfulness down. Diagnose explains why the score landed where it did. When a prior-authorization intake agent cites a payer policy its tool call never returned, Diagnose traces that to the exact turn and narrates the root cause in plain language. You get evidence and attribution instead of a vague "needs improvement," and the assessment carries its own confidence rating, so you know the difference between a clear problem and a best read that deserves a closer look. Optimize then proposes a concrete change, shown as a reviewable diff rather than a silent overwrite. A tightened prompt instruction, a tool-call guardrail, a role clarification. You accept, edit, or reject it, re-run, and compare the new scorecard against the old one. The human stays in control, and recommendations are validated by evidence rather than trust alone. That loop is what lets Citizen Developers improve an agent with confidence instead of guesswork. "We rewrote the prompt" is not an audit trail. Here is the run, here is where it broke, here is what we changed, and here is who approved it. That is an agent performance review.

Watch ↗

See what ACTAVA can do for you

Bring one workflow your team still runs by hand. We'll show you the agent that runs it.