Pasito

The outbound action path is real, but no external system it connects to is named. Campaigns send over email and SMS and automatically link members through to a microsite and decision support, so the platform writes into a member's inbox and phone and hosts the destination. Document ingestion accepts "spreadsheets, Word documents, PDFs, and more", and RFP intake management "structures RFP documents and data into a clear, navigable view", so the inbound path handles whatever a carrier or broker actually sends.

No benefits administration platform, carrier system, HRIS, payroll provider, enrollment system or identity provider is named, the site has no integrations section, and no connector directory or count is published. For a product sold to carriers, brokers and employers whose data lives in benefits admin systems and HRIS, that gap stands out. Documents and files are how data gets in, which is ingestion rather than a connected system, and serving an ecosystem of carriers, brokers and employers describes the market, not an integration.

Sourcepasito.ai/who-we-serve/employers, /who-we-serve/insurance-carriers and homepageread 2026-09-12

One stated flow runs from proposal to utilization, though the sequence is Pasito's own process rather than one the customer designs. As the company puts it, "Pasito simplifies benefits from proposal to utilization with AI agents trained to understand plans and execute workflows at scale". The stages inside that span are named individually across the audience pages: documents scanned and schemas structured, assets created, campaigns coordinated and delivered, members answered year-round.

The chain is stated as one flow rather than as separate tools: "every asset standardized, created, and delivered through AI workflows for every client", with highlight sheets, enrollment presentations and engagement communications all generated from the same normalized layer and then pushed out through linked email and SMS campaigns into a microsite and decision support. Extraction feeds generation, which feeds delivery, which feeds support.

The participants are distinct: the carrier or broker who uploads plans and reviews assets, the extraction and generation agents, the employer HR team running enrollment, the member who receives a campaign and asks the assistant a question, and the assigned success manager. A worked path runs from a carrier's plan document to an employee's coverage question being answered from it.

No orchestration runtime is named, no handoff between agents is described, and no authoring or workflow design surface appears anywhere. The sequence is Pasito's benefits process parameterized by the customer's documents rather than a flow the customer builds. Quoting and claims, which the Series A press release places in the lifecycle, are not described on Pasito's own pages.

Sourcepasito.ai/what-we-do, /who-we-serve/insurance-carriers and /who-we-serve/employersread 2026-09-12

Agents are grounded in the customer's own plan documents, which Pasito builds into a maintained, normalized store. The agentic AI "scans every document, structures new schemas, and standardizes client information for every asset", extracting "from multiple file types, including spreadsheets, Word documents, PDFs, and more". Plan documents, contracts, proposals and census files arrive unstructured and leave as a normalized store: schema inference is the product, not a preprocessing step.

Persistence is explicit. The normalized layer is described as "the same source of truth as the rest of the platform", and every downstream asset is generated from it, so it stands between runs and across products rather than being assembled per request. It scales past any context window across a carrier's whole book of plans.

Retrieval shows at the response level. Administrators configure the assistant with source documentation, and the assistant answers a member's coverage question from the employer's actual plan rather than from a model's general knowledge, which matters in a domain where a generic answer is a compliance problem. Proposal comparison and RFP intake structure documents "into a clear, navigable view", so the store is queryable by a person as well as by the agents.

The ninety-eight percent plan construction accuracy figure comes from press coverage of Pasito's funding, not from its own site.

Sourcepasito.ai/who-we-serve/insurance-carriers, /what-we-do and homepageread 2026-09-12

Admins can configure the assistant before it runs, but no approval step sits over the agent's own output. The AI benefits assistant comes with "admin controls for tone, custom instructions, and source documentation".

An administrator bounding which documents the agent may answer from is a guardrail on what it can say, and on a benefits product that is the consequential constraint: it keeps the assistant from answering a coverage question from anything but the plan the employer actually bought.

Alongside it, "clear disclosures" mean members always know AI produced the content, and the AI governance framework adds a pre-deployment validation gate.

Nothing holds the agent's answers for a person. The assistant "answers benefits questions 24/7" to employees directly, and no queue, checkpoint, confidence threshold or human sign-off is documented before a member receives an answer about their coverage. Benefits guides, microsites and comparison documents are generated for a broker or carrier to review before publishing, which is the customer looking at a draft rather than a gate built into the product. Pasito publishes the configuration and the disclosure, not the hold.

Sourcepasito.ai homepage and /securityread 2026-09-12

Security is documented in unusual detail, with an attestation whose auditor is named and whose report can be requested, and access controls stated explicitly. Pasito's SOC 2 Type II is "independently audited by Prescient Assurance, confirming our controls for security, availability, and confidentiality", with a published "Request our SOC 2 report" form and a separate audit report request route. Naming the audit firm and providing a request path goes a step past a badge-only claim.

HIPAA compliance is documented with substance rather than as a label: physical, technical and administrative safeguards for protected health information, Business Associate Agreements where applicable, and secure handling under federal regulations.

Access runs under a zero trust model: "we implement role-based access controls and least privilege principles, with routine reviews and revocation protocols". RBAC with periodic review and a revocation process covers customer facing access, and revocation protocols are rarely published at all.

The page goes further: per-client data isolation in a dedicated trust zone preventing cross-client co-mingling, AES-256 at rest and TLS 1.2+ in transit, 24/7 monitoring, regular penetration tests, vulnerability scans and static and dynamic code analysis through internal and third-party experts, background checks and confidentiality agreements on all staff and contractors, and continuous security training including threat simulations.

Sourcepasito.ai/securityread 2026-09-12

Each response carries provenance, but no record of agent conduct is published. Administrators configure the assistant with source documentation alongside tone and custom instructions, so an answer can be traced back to the plan document it came from.

Pasito commits to "clear disclosures", stating that "users are always informed when AI is used to generate or enhance content", and engagement analytics show employers and advisors where engagement is breaking down.

That is real visibility, and provenance on a benefits answer matters more than usual because a wrong answer about coverage has a cost.

Nothing retains a reconstructable record of what the agents did. No run log, action history, audit trail, export or retention period is named anywhere on the site, and nothing states that which documents the extraction agent read, what schema it inferred, what it changed, or what the assistant told a given member is captured and retrievable after the fact.

"24/7 monitoring" is Pasito's own staff watching the platform for threats, not a customer-facing trail, and "Request audit reports" is a DocuSign form for the SOC 2 report. A source citation answers where an answer came from; an audit trail answers what the agent did. Only the first is published.

Sourcepasito.ai/who-we-serve/insurance-carriers, homepage and /securityread 2026-09-12

No cross session agent memory is documented, and the security page states the opposite. Under Pasito's Zero Data Retention Agreement with Anthropic, "inputs and outputs are processed ephemerally": "the data exists only for the duration of the session and is immediately discarded once the interaction is complete", with "no data stored, logged, or used to train AI models". A vendor that sells ephemerality as a privacy feature is telling a buyer there is no cross-session agent memory, and on a product handling protected health information that is the right trade rather than a gap.

The data layer is the customer's own records: normalized plan documents, contracts and census files held as a source of truth. That is an application data model, not agent memory; it is the corpus the agents read.

Client-specific training is not memory either. Models "customized for each client using only their data" are learning absorbed into a per-tenant model, which improves the system rather than giving a customer a store to point at, and no employee data enters training "without explicit approval".

Sourcepasito.ai/security and homepageread 2026-09-12

No region, data location or hosting provider is published, so no residency option is documented. The absence sits on a security page that is otherwise precise about infrastructure: it documents encryption to the cipher and protocol version (AES-256 and TLS 1.2+), names its SOC 2 auditor and its model provider, and describes per-client data isolation, penetration testing and monitoring cadence. A page that specific about how data is protected and silent about where it sits is not concealing an option.

What looks like a customer environment is not one. "Each client's data lives in a dedicated trust zone, architected to prevent any cross-client access or data co-mingling" describes tenant isolation, a security control. A trust zone is a boundary inside Pasito's cloud, not an environment the customer selects, controls or can place. HIPAA and SOC 2 are compliance attestations, not residency statements. There is no region list, named customer environment or selection surface, and not even a one sentence statement of where data is located.

Pasito handles protected health information under HIPAA for US carriers and employers, and a buyer with a residency requirement has nothing published to point at.

Sourcepasito.ai/securityread 2026-09-12

Eight preconfigured units, each with its own page, make up a catalog arranged under three audiences in the site's menu. Benefits consultants get Benefits guides, Benefits microsites, the AI benefits assistant and Proposal comparison. Insurance carriers get Decision support, Marketing documents, the AI benefits assistant and RFP intake management. Employers get Decision support, Benefits microsite and benefits hub, the AI benefits assistant and Year-round communications. A buyer browses that menu and picks what applies, and each unit does its work once selected.

The units stand alone. Take away RFP intake management and the AI benefits assistant is still a whole working agent answering coverage questions; take away the assistant and proposal comparison still structures a broker's RFP documents into a navigable view. These do different jobs for different people, not three modes of one engine.

They ship configured rather than built. Pasito's own framing is "AI agents trained to understand plans and execute workflows at scale" and "custom AI agents and workflows", with assets "templatized" and every asset "standardized, created, and delivered through AI workflows for every client". Templates plus prebuilt agents arrive working rather than requiring authoring.

The units are vertical specific by design, and there is no library a customer adds to, no install path and no third-party contribution route. What is cataloged is Pasito's own product surface. White-label branding of these agents is presentation, not extensibility.

Sourcepasito.ai/what-we-do, /who-we-serve navigation and /who-we-serve/insurance-carriersread 2026-09-12

Channels are named, schedules are sold as a product, and conversations open off data conditions. Pasito's employer pages say customers can "build and launch email and SMS benefits communications campaigns that automatically link to your benefits microsite and decision support".

Email and SMS are two named outbound channels; the benefits microsite and benefits hub is a third, persistent surface members return to; and the AI benefits assistant is a fourth, conversational one that "answers benefits questions 24/7".

Campaigns are coordinated so a send lands a member on the microsite and into decision support, a channel chain rather than four separate sends.

Schedules are a named product. Year-round communications is its own page and product line, distinct from open enrollment campaigns, so cadence is something the customer configures across a benefits year rather than a one-off blast. The assistant runs 24/7 with no operator needed to start it.

Events drive the proactive half. Decision support is described as "proactive and personalized", leading with a recommendation and "removing lengthy intake forms", and identifies best-fit plans from preset demographic information, so the agent opens the conversation off a data condition rather than waiting to be asked. The enrollment cycle is the recurring event the whole product keys on. Events, schedules and multiple channels are all documented.

Only two outbound channels are named. No voice, no chat platform such as Teams or Slack, and no WhatsApp appears anywhere, a narrow set for a product reaching employees.

Sourcepasito.ai/who-we-serve/employers, /employers/decision-support and /employers/year-round-communicationsread 2026-09-12

Pasito names a single model provider and offers no customer choice. Its security page states, under the heading "Zero data retention with Anthropic", that "Pasito maintains a Zero Data Retention Agreement with Anthropic, the AI provider behind Claude". One provider is identified by name with a contractual term disclosed alongside it.

No selection surface exists: no model picker, admin setting, per agent assignment or bring your own key path is published. The admin controls that do ship govern tone, custom instructions and source documentation, which is what the agent says and reads, not what runs it.

Client-specific training is not model flexibility. "Models are customized for each client using only their data and Pasito's generalized training data, never shared or repurposed" describes tuning behavior to a customer, not letting the customer choose a provider, and the plural refers to those tuned variants rather than a choice set. Strict consent policies gate whether employee data enters training at all. Pasito chooses the model and says so openly, contract term and all.

Sourcepasito.ai/security and homepageread 2026-09-12

No developer surface is published: no API reference, endpoint list, authentication guide, SDK, webhook or MCP server, and no developer, docs, API or integrations section on the site. app.pasito.ai is the application itself behind a login, and there is no docs or dev subdomain, llms.txt or page for assistants.

White labeling is not extensibility. Deploying an agent under a client's own brand is presentation (the logo, the tone, the domain) and says nothing about whether an outside system can call Pasito. No shipped capability lets clients build their own agents in the workspace.

The product shape is consistent with the absence. Pasito is sold through demos into carriers, brokers and employers, with a success manager assigned to every client, and there is no self-serve tier a developer could reach. A vendor delivering configured agents as a managed workspace has little reason to publish a build surface.

Sourcepasito.ai/security, homepage navigation and footerread 2026-09-12

An AI governance framework tests changes before launch and evaluates them after. Under the heading "Generative AI use policy highlights", Pasito's security page states that "pre-deployment testing validates AI output accuracy, data integrity, and ethical alignment before launch". A change is tested before release, with accuracy named as the thing measured.

After launch, under "Post-launch monitoring", "ongoing evaluations ensure AI tools meet performance, fairness, and safety standards", a controlled loop against stated standards rather than a live dashboard. An incident response protocol closes it, with the data team investigating and resolving deviations, and a feedback loop lets clients and users submit concerns directly, feeding into platform updates and enhancements.

No scoring scale, test set construction, pass threshold or readable verdict format is published, so a customer cannot see the result of a validation, only that one occurred. The harness is Pasito's own, gating Pasito's releases, and no customer-run sandbox for testing a changed agent before it reaches members is documented. Bias minimization ("we follow industry AI ethics guidelines") is a guideline commitment rather than an audit, though fairness does appear among the post launch evaluation standards.

The reported plan construction accuracy of ninety-eight percent, against an industry average of seventy-four, appears in coverage of Pasito's funding rather than on the vendor's own pages.

Sourcepasito.ai/securityread 2026-09-12

The work here is document processing and content delivery, and neither touches an interface, so the product's shape rules out browser and computer use rather than merely omitting it. Pasito's agentic AI "scans every document, structures new schemas, and standardizes client information", extracting from spreadsheets, Word documents and PDFs. Assets are generated from that store, campaigns go out over email and SMS, and members read a microsite and ask a conversational assistant. Every step is file parsing, generation or messaging.

Nothing navigates. No hosted or local browser, desktop session, remote or local computer control, RPA or browser extension appears anywhere on the site. Pasito's inputs arrive as documents a client uploads rather than as pages someone browses.

Asking how often Pasito breaks when a target application changes its on screen layout means nothing for this product. A PDF's layout changing is a parsing problem, not a UI automation one, and the vendor addresses it by inferring schemas rather than by clicking. The benefits microsites and benefits hub are surfaces Pasito builds and hosts for members to visit, which is publishing rather than operating someone else's interface, and white-labeled deployment of agents under a client's brand is likewise presentation.

Sourcepasito.ai/what-we-do, /who-we-serve/insurance-carriers and homepageread 2026-09-12