My AskAI

Agents act in outside systems, including consequential actions: AI Tasks are sold on the Pro plan with order refunds given as the worked example, alongside live user data connections pulling recent orders from back-end systems, AI tagging writing contact reasons back into the helpdesk, and intelligent escalation transferring a conversation to a named human inside the customer's own tool. Breadth spans six helpdesks with native apps in their marketplaces, Zendesk Support and Messaging, Intercom, Freshdesk, Freshchat, HubSpot and Gorgias, plus Shopify, Salesforce and eleven knowledge sources.

Sourcemyaskai.com/pricingread 2026-09-05

The agent resolves a question end to end inside one conversation: retrieve from the vector store, answer in the customer's language, run a task such as an order refund where one is configured, tag the contact reason, and escalate to a person with the thread intact when it cannot answer. This is one bounded support flow rather than multi-step or multi-agent execution.

Tasks and Tools are individual configured actions rather than chained sequences, and no branching runtime, workflow builder or agent-to-agent handoff is documented; multi-agent setup on the Scale plan means separate agents per brand or region, not agents co-operating on a case.

Sourcemyaskai.com/features/tasks-and-toolsread 2026-09-05

A maintained retrieval structure documented down to the store: uploaded content is embedded as vectors and held encrypted on Qdrant, a vector database, alongside reference content carrying text, author and links.

Around it sit eleven-plus knowledge sources synced and kept up to date automatically, including Zendesk, Intercom, Freshdesk, Gitbook, Confluence, Google Drive, Notion, SharePoint, Dropbox, Salesforce, OneDrive and any public website, plus unlimited help articles and documents, up to ten websites on Scale, training on historic tickets, and self-learning. The store persists, scales past the context window and stays queryable.

Sourcesupport.myaskai.com/security-+-privacy/securityread 2026-09-05

Two mechanisms, neither an approval gate the vendor owns. Copilot mode has the agent draft a reply as an internal note that a human edits and sends, so a person is in the path by construction, and internal knowledge sources can be restricted to copilot only so they never reach a customer.

Intelligent escalation offers the customer a talk to a person option and transfers them to a named agent inside the existing helpdesk, and Guidance constrains tone and behavior.

Where a human approves, they approve inside Zendesk or Intercom, and that gate belongs to the helpdesk rather than to My AskAI; no pre-send approval surface, sign-off step or policy threshold of its own is documented.

Sourcemyaskai.com/features/copilot-internal-note-replyread 2026-09-05

Compliance is documented; access control is not.

The documented side includes SOC 2 Type II, a trust center at trust.myaskai.com, AES-256 encryption at rest and TLS in transit, uploaded content held in isolated containers, files automatically deleted after embedding unless the customer opts out, a commitment that data is never used for anything but servicing API calls or requested support including model training, and GDPR documentation covering processors, retention and architecture.

SOC 2 Type II is sold from the Scale plan upward. No identity integration, named role model or permission scheme is documented; the plan matrix sells team seats and unlimited seats without describing what a seat can do.

Sourcesupport.myaskai.com/security-+-privacy/securityread 2026-09-05

Insights generate conversational analysis across every customer interaction, tagging labels contact reasons, usage and resolution are tracked per plan, alerting fires when a new topic is identified or an existing one spikes in a short window, and conversations and insights export. This reports what customers asked and how often, not what the agent did. No step-level decision trace, retrieval log naming which source produced an answer, or configuration audit trail is documented.

Sourcemyaskai.com/features/insightsread 2026-09-05

Memory with a stated scope or lifetime is not documented. What persists is the knowledge store, embedded vectors held on Qdrant, and training on historic tickets, which is corpus rather than retained agent state. The security documentation describes uploaded files being deleted after embedding and data used only to service API calls, which describes state deliberately not kept.

Sourcesupport.myaskai.com/security-+-privacy/securityread 2026-09-05

A named single region, stated precisely in the vendor's own security documentation: uploaded content is embedded as vectors and stored encrypted on Qdrant, run on Google Cloud Platform and located in The Dalles, Oregon, in us-west1-gcp. One region, named, with no menu and nothing for a buyer with a different requirement to select, and no VPC, single-tenant or self-hosted option anywhere. Multi-agent setup across brands, regions or customer types, sold from the Scale plan, is customer segmentation rather than data residency.

Sourcesupport.myaskai.com/security-+-privacy/securityread 2026-09-05

No prebuilt agent, template gallery, cloneable starter or marketplace of packaged assets appears anywhere on the site, whose features index runs to capabilities rather than assets. The agent is built from the customer's own content, help centers, documents, websites and historic tickets, with Guidance and custom answers layered on top, all authored by the customer rather than adopted from a set. Multi-agent setup on Scale and unlimited agents on Enterprise create more instances of the same configurable agent for different brands or regions, not a library to choose from.

Sourcemyaskai.com/pricingread 2026-09-05

Conversations arrive through inbound helpdesk queues rather than a surface My AskAI owns: native apps installed in Zendesk Support and Messaging, Intercom, Freshdesk, Freshchat, HubSpot and Gorgias handle new conversations before a human sees them, alongside email and ticket replies, a standalone widget, a Chrome extension copilot for agents, and Slack and Teams for internal questions, in 95-plus languages with automatic detection.

Alerting fires on newly identified or spiking topics. Text only, with no voice channel documented anywhere, and no scheduled, webhook or event-driven trigger surface beyond conversation arrival.

Sourcemyaskai.com/pricingread 2026-09-05

A single provider named in the vendor's own security documentation: uploaded files are added to the AI support agent by being uploaded to OpenAI, and files are then automatically deleted unless the customer says otherwise. No model selection, routing or bring-your-own-model control appears anywhere in the plan matrix or the feature set.

Sourcesupport.myaskai.com/security-+-privacy/securityread 2026-09-05

A documented API for My AskAI's own platform: the Query API publishes its endpoint, API-key authentication, request schema, a 750-character limit with the exact error returned when it is exceeded, and response fields including unknown_answer and human_handover, sitting in a dedicated API documentation section of the help docs. Alongside it the vendor ships an MCP server, linked from the site footer, and a Customer Data API for personalizing responses with live data such as recent orders. API, Slack and Teams access is a $49/mo add-on on Pro and included from Scale.

Sourcesupport.myaskai.com/api-documentation/query-apiread 2026-09-05

A quality gate in the release path plus an optimization loop after it. Testing and QA ships as a named feature, and replies can be trialed privately as internal notes inside the helpdesk before the agent is turned customer-facing, so a change is exercised against real tickets without a customer seeing it.

After deployment, the Improve surface identifies knowledge gaps and proposes ways to close them, self-learning drafts new answers from past tickets, and resolution rate is the measured outcome, with customers reporting movement from first sync to a higher deflection rate by following the suggestions. What the Testing and QA feature covers beyond internal note trials is not published.

Sourcemyaskai.com/features/testing-and-qaread 2026-09-05

No browser control, hosted session, virtual desktop or computer use appears anywhere. The agent works inside the customer's existing helpdesk through installed apps and reaches other systems through its Query API, Customer Data API and configured Tasks, so the route out is programmatic by the vendor's own description rather than unstated. The Chrome extension is an agent-facing copilot surfacing draft replies to a person, which displays in an interface rather than operating it.

Sourcemyaskai.com/pricingread 2026-09-05