Outverse

Writes into the customer's systems are authenticated and shown end to end, not just claimed. Outverse sits between the customer's channels, systems and service logic, works with the platforms, internal systems and tools a team already relies on, and lets teams connect their own tools and custom actions. The actions matter. The published agent trace verifies a payment ID, checks payment status and replies, and the running policy view steps through requesting and verifying customer information, processing a subscription change, creating a deal and confirming resolution.

The voice product connects directly into the customer's CCaaS platform while the helpdesk stays the system of record. At Jeeves, Outverse connects securely to internal tools to verify customer identities and retrieve live data without exposing sensitive systems or needing engineers, and every automated action leaves a record in Zendesk as well as in Outverse. There is no integration directory or named connector catalog, so the breadth of prebuilt connections is unknown, even though the mechanism and the write capability are shown.

Sourceoutverse.com homepage, voice page and Jeeves case studyread 2026-09-05

Multi step service work is composed in natural language and executed with tool calls interleaved. Customers define intents so agents classify and route, then write agent SOPs that orchestrate behavior and workflows, and Outverse sits between channels, systems and service logic so agents work across real service workflows. Execution runs in sequence and keeps state, not as a single reply.

The running policy view shows an agent at step three of five, moving through requesting and verifying customer information, processing a subscription change, creating a deal and confirming resolution, and the published trace interleaves intent recognition, policy selection and successive tool calls before replying.

Tickets carry in progress, answered and escalated states with elapsed time, and deployment spans channels from a unified orchestration center. On voice the same orchestration runs inside a single call, verifying identity, looking up account data, applying policy and acting.

Policy driven execution and failure pattern tracking imply branching, retry and failure handling, but no condition, fallback or retry construct is described on its own.

Sourceoutverse.com homepage and voice pagesread 2026-09-05

Agents answer from the customer's own material, but Outverse does not say how that material is held. Agents work against intents the customer defines and natural language SOPs and policies, the voice product is shown against a customer's own call flows and support materials, and the published example shows the agent following a named policy for payment status requests before calling tools and replying. Grounding in customer content is real and is the basis of the product.

There is no ingestion, indexing, retrieval, chunking, refresh or source management surface, and nothing says whether a maintained corpus exists or policies are read at answer time. Outverse launched in 2021 as a community and knowledge base platform with forums, documentation and semantic search, and material about that generation still circulates, but it is a different product from the service agent platform described here.

Sourceoutverse.com homepage and voice pagesread 2026-09-05

Behavior is bounded by policy the customer writes, and escalation is a designed state, not a fallback. Operations teams define customer intents so agents classify and act accordingly, then write agent SOPs and policies in natural language that orchestrate behavior, including handoffs. Policies and intents are the first component of the product, which defines logic, policies and handoffs.

The customer owns this written control layer, and operations teams govern it without every change becoming an IT project. Guardrails are a named product property, with agents described as guardrailed and holding up in real operations, and on voice, behavior is refined within defined guardrails as coverage expands.

Handoff is visible in the product, since the activity view carries escalated as a distinct ticket state alongside answered and in progress, and the published policy run ends by confirming resolution with the customer. Style guidelines and business guardrails are named alongside the compliance section.

There is no approval queue before sending, draft review mode or sign off for each action, so oversight is written in advance and inspected afterwards, not exercised message by message.

Sourceoutverse.com homepage and voice pagesread 2026-09-05

Certification is well evidenced, but controls for customers are not. Outverse states it is SOC 2 and ISO 27001 compliant, shows the AICPA SOC and ISO 27001 marks in its security section, and runs a trust center at trust.outverse.com on Vanta's EU instance.

Data handling claims sit alongside it, with encryption at rest and in transit and a commitment that customer data is never used to train AI models. There is no SSO or SAML, role based access control, admin audit log or configurable retention.

The privacy policy does list access controls, encryption in transit, logging and backups, but that is the generic security paragraph every processor publishes, not a control a customer configures.

Sourceoutverse.com homepage security section and privacy policy section 7, trust.outverse.comread 2026-09-05

Teams can reconstruct the agent's reasoning, and Outverse names this as one of the four components of the product. Teams can inspect what an agent saw, what it decided, which tools it used and where it handed off, and the governance layer shows decisions, tool use, workflow paths and handoffs in detail. The published trace works at that grain, showing the understood intent, the policy being followed, each tool call in sequence and the reply, and the running policy view shows the step reached within a multi step policy.

The activity view lists tickets with status, channel, response and elapsed time, with escalated as a distinct final state. On voice the same holds for each call, with ops teams monitoring behavior and understanding the reasoning behind decisions. At Jeeves, every automated action leaves an auditable record in both Zendesk and Outverse, and the customer credits that visibility for keeping trust and compliance. Coverage, outcome and failure pattern tracking sits in the same analytics component.

Sourceoutverse.com homepage, voice page and Jeeves case studyread 2026-09-05

An Outverse agent holds state across a single interaction. The published trace runs a multi turn exchange in which the agent identifies intent, asks for a payment ID, receives it, verifies it, checks status and answers, carrying the reference forward between turns, and the running policy view shows an agent at step three of five with the earlier steps done. On voice the same holds inside one call, where the agent verifies identity, looks up account data, applies policy and acts before the call ends.

That is conversation state inside a single session, which every chat interface has. There is no memory layer with a stated scope and lifetime. Nothing says what an agent keeps about a customer between separate contacts, whether memory is scoped per user, team or workflow, how long it lives, or whether an operator can inspect or delete it. Where continuity exists across contacts, it belongs to the helpdesk, which Outverse says stays the system of record.

Sourceoutverse.com homepage and voice pagesread 2026-09-05

Data location is explicitly not constrained. The privacy policy says Outverse may host and process personal data in the UK, the EU, the US or other regions, relying on lawful transfer mechanisms when data crosses borders. That is the opposite of a residency commitment, since no region is guaranteed, none is named as primary, and the customer is told to expect movement instead of being offered a choice.

There is no self hosted, on premises, private cloud, single tenant or VPC option, and no region selector or residency configuration. Mentions of multiple regions elsewhere describe where the customer's support operation runs, not where its data is held. Residency commitments can sit in an enterprise contract without appearing on a marketing page.

Sourceoutverse.com privacy policy section 6read 2026-09-05

Assets exist, and the customer assembles them. Agents are composed from two written artifacts, customer intents defined flexibly so agents classify and act accordingly, and agent SOPs and policies written in natural language to orchestrate behavior and handoffs. Policies and intents are the first component of the deployment workstack, so they are core objects in the product, not prompt text, and a running policy is shown at step three of five.

The customer writes them, though, and nothing is supplied ready to adopt. There is no template gallery, starter policy set, named prebuilt agent, bundled toolkit or entitlement, and the route to a first deployment is analysis, not selection, with Outverse working from the customer's past ticket data to find automation opportunities and scope a proof of value.

Sourceoutverse.com homepageread 2026-09-05

Channel breadth is real, but work only comes in one direction. Voice is a real channel, not a roadmap item. Outverse deploys across email, chat, voice and APIs from what it calls a unified orchestration center, and ships a low latency voice agent that connects directly into the customer's CCaaS platform, verifies identity, looks up account data, applies policy and acts within a single call. Deployments run across multiple regions and languages.

Every path begins with an inbound contact. The published surfaces open on a user question or an incoming call, the activity view is a ticket queue, and there is no scheduled run, outbound campaign, proactive follow up, webhook trigger or event subscription. APIs are named as a deployment channel, which is a way in, not a trigger framework, and no event catalog comes with it.

Sourceoutverse.com homepage and voice pagesread 2026-09-05

Outverse says little about models. The only statement for customers is that teams can apply deeper model and data controls where required, listed among configurable controls in the governance section, alongside a claim that teams benefit from advances in AI without reworking their workflows, policies and integrations, which describes a model layer Outverse manages and upgrades under a stable configuration.

Together these put model handling inside the platform, with some adjustment open to the customer. No model provider is named, there is no selector, routing rule, assignment per workflow or way to bring your own key, and the phrase about deeper controls names no mechanism, so what a customer can actually set is unknown. The privacy policy names no AI subprocessor, and the only related commitment is that customer data is never used to train AI models.

Sourceoutverse.com homepage governance and security sections, privacy policyread 2026-09-05

Outverse names APIs and SDKs as a product component but publishes nothing behind them. It lists channels, SDKs and actions as one of the four parts of its deployment workstack, covering deployment across email, chat, voice and APIs and connecting the customer's own tools and custom actions, so Outverse itself asserts an API surface and SDKs.

The terms of service back up that a programmatic surface exists, since they prohibit attempts to bypass security or rate limits, language written for API users. There is no endpoint list, authentication model, SDK package, client library, versioning statement, webhook catalog or MCP server, and no developer portal or docs subdomain. The product is sold through sales, with the app behind a login at app.outverse.com.

Sourceoutverse.com homepage, terms and conditionsread 2026-09-05

Customers gate changes before deployment and read results afterwards, with their own team running both. Test, iterate and deploy is the third step of building an agent, covering testing agent performance and making safe updates to keep managing and improving results, and Jeeves tests policy updates safely before deployment. The analytics component tracks coverage, outcomes, failure patterns and improvement opportunities over time, a readable, comparable measure of whether the agent resolves what it was meant to resolve, not a pass or fail on a single run.

Voice carries the same loop, with ops teams refining performance within defined guardrails as coverage expands. The whole product is positioned on this, a governed layer where operations teams manage logic, performance and workflow changes without every update becoming an IT project. There is no regression suite, versioned test set, scored evaluation run or published benchmark, so this is a safe change workflow plus operational measurement, not an offline evaluation harness.

Sourceoutverse.com homepage and voice pages, Jeeves case studyread 2026-09-05

Browser control, hosted sessions and interface automation appear nowhere, and Outverse states the alternative plainly. It reaches systems through connected tools and custom actions, and every action shown is a tool call. The published trace calls verify payment ID and check payment status, and the voice product connects directly into the customer's CCaaS platform and the systems the team relies on.

At Jeeves, the connection to internal tools is made securely without exposing sensitive systems or needing engineers, which is an integration, not an agent driving a screen. None of the product's components, policies and intents, reasoning and observability, channels and actions, and analytics, names a browser, a virtual desktop or a login based path into an interface.

Sourceoutverse.com homepage and voice pages, Jeeves case studyread 2026-09-05