Konvo

Authenticated writes into systems of record, with the actions enumerated. Konvo publishes its skill catalogue by name: get order status, cancel order, edit order shipping address and refund order under order management, manage subscription deliveries and edit subscription shipping address under subscriptions, and close conversation, escalate to an agent and start a side conversation as general actions. Those are state changes in the merchant's commerce stack, not lookups.

The connected systems are named too, spanning shop platforms with Shopify and WooCommerce, fulfilment and logistics with Mintsoft, Amphora and Sendcloud, helpdesks with Gorgias, Zendesk, Front and Freshdesk, and marketing and CRM with Klaviyo and Salesforce, which the vendor summarises as connecting a merchant's 3PL, shop system and CRM.

Beyond structured connectors the agent is described automating tedious actions like calling a courier or emailing a provider, so it reaches parties that have no integration at all. Re-based rather than regraded, and the sourcing is now first party: the prior Full cited Crunchbase materials alongside the vendor. Recorded as the boundary: no connector directory or count is published, so the named set above is the evidence and a buyer should confirm their own stack.

Sourcekonvoai.com AI platform section and FAQread 2026-09-05

Multi-step processes that reach outside the platform to finish. Processes is a first-class build component alongside Knowledge, Persona, Skills and Recommendations, and the vendor's own framing is that its agents act, reason and behave like a human, following complex processes and executing actions such as order changes, carrier communication and internal processes while holding the brand's tone.

The published examples are sequences rather than replies: resolve a cancellation, refund or update by reading the shop system and writing back to it; handle a courier incident by calling a courier or emailing a provider, which means the run continues through a third party before it can close; manage a subscription by adjusting deliveries.

Skills compose into those processes and include branching outcomes, since escalate to an agent, close conversation and start a side conversation are all terminal or forking steps the agent selects between. The vendor contrasts this with helpdesks that add AI as a feature, positioning the agent as doing the work rather than assisting it. Re-based rather than regraded, and the citation is now first party rather than the vendor's site read alongside Crunchbase. Recorded as the boundary: no runtime, step model, retry or failure handling is individually documented, and the flow builder detail sits in a help centre that is machine refused.

Sourcekonvoai.com AI platform section and FAQread 2026-09-05

A named, indexed corpus with its own management surface. Konvo's Knowledge hub is one of five build components alongside Persona, Skills, Processes and Recommendations, and the published view shows it holding 49 sources with a search box across them: a crawled storefront at 23 pages, a crawled FAQ site at 5 pages, an uploaded order-management protocols document of 355 KB, and a countries-we-deliver-to entry, with an Indexing status shown against a source while it is processed.

The vendor describes the model plainly, that Konvo learns like a human teammate from a connected website, docs, past conversations and processes. Crawl, upload, index, search and per-source status is a maintained retrieval structure that persists and stays queryable, which is the published line, and persistence rather than external sourcing is the discriminator.

Corrects the reasoning while holding the grade, and the correction matters: the prior Full rested on agents accessing real-time customer, order and catalogue data within connected systems, which is context assembled per run from live integrations and is Partial under the published bar, so the cell was right by accident. It also cited Crunchbase materials alongside the vendor, and an aggregator profile is not evidence of capability.

Sourcekonvoai.com AI platform section, Knowledge hubread 2026-09-05

Escalation is designed in and approval is not. Konvo's framing is that it handles every conversation and the team steps in only when it matters most, with the agent resolving end to end or escalating when needed, alerting staff when it needs more information to resolve a query, and handing conversations to the human team when it needs help.

The published example shows the shape: a customer asks whether a product is safe with a nut allergy, the agent replies that it is looping in a teammate, and the inbox flags that the agent needs help, with full context kept across every channel and agent. Escalate to an agent is one of the configurable general skills alongside close conversation and start a side conversation, so an operator decides when the agent hands off.

Limitation, which is the Full requirement: the published bar asks for the vendor's own review and approve surface, and nothing gates what the agent sends. No pre-send queue, draft-review mode, confidence threshold or per-action sign-off is documented, and the agent autonomously cancels orders, issues refunds and edits shipping addresses once a skill is enabled.

Corrects a Full resting on alerting and handover, which is escalation, the shape graded Partial elsewhere in this lane, and which cited Crunchbase materials alongside the vendor. The underlying facts hold up first party and they do not reach the bar.

Sourcekonvoai.com AI platform and FAQ sectionsread 2026-09-05

Neither half of this axis is documented. Compliance posture: no certification or attestation is claimed anywhere on the site. No SOC 2, ISO 27001 or equivalent appears on the homepage, the pricing page, the customers pages or the footer, and no trust centre or security page exists on any subdomain.

What Konvo does publish is a European legal tier appropriate to a German company, with an imprint, a privacy policy, terms and conditions and a data processing addendum linked from every page, which establishes GDPR-facing obligations without amounting to an attestation. Access surface: nothing documents who inside a merchant's organisation can do what. Agents and teams exist, since conversations are assigned and claimed, but no role model, permission set, SSO or SAML is published.

Probed as the bar requires before concluding: site navigation and footer, the legal tier, a domain-scoped search for security terms, and the help centre, which is machine refused. Under the published bar either half alone carries Partial and neither is present here. Re-based off the vendor's own pages; grade unchanged and the cell now records what was searched and what the legal tier does establish, rather than resting on nothing documented in retrieved materials. Recorded as unexhausted: docs.konvoai.com returns a bot-detection block, and admin documentation is where an access model usually lives.

Sourcekonvoai.com footer legal tier and site navigationread 2026-09-05

Insight into the operation, not a record of the run. Analyze is a named stage of Konvo's own lifecycle with a Dashboard behind it, and the vendor describes what it produces: uncovering opportunities, flagging missing information so the agent keeps improving, and reporting agent performance and customer behaviour through AI-driven insights. Headline operating figures are published, 2.9 million AI conversations exchanged and a 70 percent average resolution rate.

So a merchant can see how the agent is performing and where its knowledge is thin. Limitation, which is the Full requirement: the published bar says reporting is not auditing and seeing what happened is not reconstructing why. Nothing documents inspecting an individual run, which knowledge-hub source grounded a reply, which skill fired, what was written back to Shopify or the 3PL, or why the agent chose a refund over an exchange.

No action-level trace, decision log or audit surface appears on any page reached. That gap carries weight here because the agent cancels orders and issues refunds in the merchant's systems. Corrects a Full that read AI-driven insights as first-class visibility, which counts the reporting half as the whole axis, and which cited Crunchbase materials alongside the vendor.

Sourcekonvoai.com AI platform sectionread 2026-09-05

Context carries across the conversation and across channels, and no memory layer is described. Konvo states that full context is kept across every channel and agent, so a thread that starts on Instagram and continues on WhatsApp does not restart, and that recommendations draw on context, intent and past behaviour, which implies something retained about a shopper beyond the current exchange. Past conversations are also named among the sources the agent learns from at build time.

Limitation, which is the Full requirement: the bar asks for a documented memory layer with a stated scope and a stated lifetime, and neither is given. Nothing says what the agent retains about an individual customer between separate contacts, whether memory is scoped per shopper, per order or per channel, how long it lives, or how an operator inspects or clears it, and the published operational test cannot be answered, since a buyer could not say where the memory lives or delete it for one shopper.

Under the one-fact rule the past conversations ingested into the Knowledge hub are spent on Knowledge grounding, where they carry the Full, and a corpus is not a memory layer. Re-based rather than regraded. The prior Partial inferred retained context from the platform re-engaging past buyers and tracking interactions, using the word implying and citing Crunchbase materials; the conclusion survives on Konvo's own pages.

Sourcekonvoai.com AI platform sectionread 2026-09-05

No deployment option of any kind is documented. Konvo is delivered as a hosted service reached through a web app, distributed heavily as a Shopify app and embedded into a merchant's existing helpdesk, and no page reached names a region, a region selector, a VPC, a single-tenant arrangement, an on-premises build or a customer-tenant deployment.

The hosting location is not disclosed either, so this is not even the named-single-region case the axis treats as its best-evidenced None; the only jurisdictional signal is that Konvo GmbH is a German company publishing a European legal tier, which is a fact about the seller rather than an offering about where data sits.

Under the published bar the test is whether a buyer with a different requirement could get a different answer, and nothing here offers one. Re-based off the vendor's own pages; grade unchanged and the cell now states the delivery shape affirmatively rather than resting on a silence.

Recorded as adjacent and not credited: the data processing addendum in the footer governs how personal data is handled, which is a legal instrument rather than a deployment choice.

Sourcekonvoai.com homepage, pricing and footerread 2026-09-05

Prebuilt skills a merchant switches on rather than builds. Konvo ships a named catalogue of ready actions grouped by domain: order management with get order status, cancel order, edit order shipping address and refund order; subscription management with manage subscription deliveries and edit subscription shipping address; and general handling with close conversation, escalate to an agent and start a side conversation.

A merchant adopts these as delivered, which is why the vendor can promise most customers are live in under a week, and Recommendations arrives the same way, described as Konvo learning the catalogue and guiding customers to the right products without the merchant building a recommender. Under the published bar Full is customer-adoptable packaged assets by any route, a browsable catalogue being strong evidence rather than the definition, and a named set of prebuilt capabilities delivered ready to enable satisfies it.

Re-based rather than regraded, and the reasoning is now the skills catalogue rather than the phrase productized eCommerce agent capabilities, which described the product without testing the bar. Recorded as the boundary and the reason confidence is not higher: these are skills of one agent rather than separate agents, so under the Q32 narrowing they would not survive the test of removing one and asking whether the others are whole products. What carries the cell is that they arrive prebuilt and adoptable, not that they are a product line.

Sourcekonvoai.com AI platform sectionread 2026-09-05

Six separate inbound queues plus documented outbound initiation. Konvo states its agent works across all a merchant's channels and names them: live chat, email, WhatsApp Business, Instagram DMs and Facebook Messenger, with voice added in the deploy section, and stresses that it is the same intelligence, personality and resolution capability at every touchpoint rather than a thinner bot on the smaller channels.

Deployment runs two ways, through Konvo's own multichannel inbox or by bringing its capabilities into an existing helpdesk, where Konvo operates as another agent inside Gorgias, Zendesk, Front or Freshdesk, so work also reaches it through the queue the merchant already runs.

Outbound is documented too, on Konvo's own Shopify App Store listing, which names Broadcasts and Flows as AI-powered campaigns and automations and describes re-engaging past buyers through automated marketing campaigns, alongside unlimited no-code flow automations.

That listing is written and submitted by the vendor and is first party for what it claims, with the channel named here because a listing documents the product as sold through that channel; the capability it establishes is that the agent initiates as well as receives. Multiple inbound queues and outbound campaign initiation are two of the things the published bar looks for, and this is breadth of arrival path rather than breadth of surface. Re-based off the vendor's own pages.

Recorded as the boundary: no event catalogue, schedule or webhook trigger is enumerated on any first-party surface reached, and the flow-level trigger detail sits in a help centre that is machine refused, so the initiating events themselves are named only in general terms.

Sourcekonvoai.com AI platform and FAQ sections, Konvo Shopify App Store listingread 2026-09-05

No model provider, selector or routing rule appears on any page reached. The agent's configurable surfaces are published and models are not among them: Knowledge, Persona, Skills, Processes and Recommendations are the five build components, with Persona covering personality and tone of voice rather than the model underneath. No provider is named anywhere, no version is stated, and no bring-your-own-key path or per-skill assignment is offered.

Nor is there evidence of vendor-internal routing across providers, which is the shape the bar places at Partial. The positioning is consistent with the model being an implementation detail Konvo holds, since the product is sold on being onboarded in minutes with a Conversational AI Expert doing the tuning. Under the published bar this axis measures who chooses, customer or admin control is the Partial to Full line, and provider undisclosed is None.

Re-based off the vendor's own pages; grade unchanged and the cell now names the configurable surfaces that exist and observes that models are not among them, rather than resting on nothing documented in retrieved materials. Recorded as unexhausted: the help centre is machine refused and its AI configuration section, described as nine articles, is where a model setting would appear if one exists.

Sourcekonvoai.com AI platform sectionread 2026-09-05

No API, SDK, developer documentation or MCP endpoint for Konvo's own platform appears on any page reached. The site has no developer section; its links are the AI platform, customers, pricing, a help centre, a sign-in and a WhatsApp link generator.

Every documented interface runs inward, with Konvo connecting to a merchant's shop system, WMS, 3PL, CRM and helpdesk, and the vendor's stated position is that Konvo operates as another agent inside Gorgias, Zendesk, Front or Freshdesk while their reports, metrics and rules stay in place, which is Konvo integrating into someone else's platform rather than exposing its own. That is credited on Integrations. Under the published bar Full requires a documented API or SDK for the vendor's own platform, and nothing documented is None.

Corrects a Partial reading that drag-and-drop flow building and broad platform integration indicate a configuration and connection surface short of a documented public API. That is an inference from adjacent facts rather than a documented extension mechanism, the word indicating did the work, and the basis cited Crunchbase materials alongside the vendor. Flow building is orchestration and platform connection is integration; neither is this axis.

Recorded as unexhausted and it is the cell most likely to move: Konvo's help centre at docs.konvoai.com is machine refused, returning a bot-detection block rather than content, and an API reference would live there.

Sourcekonvoai.com homepage and FAQread 2026-09-05

A test suite run before deploy, named as a stage of the product's own lifecycle. Konvo lays the platform out as Build, Evaluate, Launch and Analyze, with Test as the whole of the Evaluate stage, and describes it directly: simulate customer conversations to see how the agent handles real scenarios, define typical conversations and run your full test suite automatically to deploy with confidence, and give feedback in the chat so the agent improves.

That is two of the four published Full routes at once, scored test cases against a defined set and a quality gate sitting in a release path, and the change under test is the customer's own agent configuration before it reaches a buyer. The onboarding motion reinforces it, with a Conversational AI Expert assigned to optimise results over time.

Corrects a Partial that credited a test panel in the AI Persona builder and an assigned expert but called it short of a documented full simulation or eval suite; the vendor documents exactly that on its homepage, using the words test suite. Recorded as the boundary: no pass rate, score or published threshold is attached to a suite run, so this is a rehearsal harness rather than a measured one, and no post-deployment holdout or controlled experiment is offered.

Sourcekonvoai.com AI platform sectionread 2026-09-05

No browser, desktop or remote computer control appears anywhere, and the three modalities are the bar. Konvo reaches external systems through credentialed integrations with shop platforms, 3PLs, CRMs and helpdesks, and its published skills are structured API-shaped actions such as cancel order, refund order and edit shipping address rather than interface operations. Its own surfaces are a multichannel inbox and an embed inside the merchant's helpdesk, both of which the bar excludes explicitly.

A platform described as connecting to a merchant's shop system, WMS, 3PL and CRM is the positive statement the bar places at None rather than a silence.

Recorded because it is adjacent and refused deliberately: Konvo describes the agent automating tedious actions like calling a courier or emailing a provider, which is real action reaching a third party that has no integration, but a telephone call and an email are not a browser, a desktop session or remote computer control, and the three modalities are the bar rather than a rationale to reason past. That capability belongs in the vendor prose, which is the same treatment ruled for IVR keypad navigation.

Under the standing rule, nothing documented is None. Re-based off the vendor's own pages; grade unchanged and the cell now states the affirmative direction and names the near miss.

Sourcekonvoai.com AI platform sectionread 2026-09-05