Tellius

Kaiya is an MCP client that queries connected servers and writes back where a server supports it, with Snowflake supporting reads and writes, and agents trigger next best actions, generate target lists and send briefings across sources such as Salesforce, HubSpot, NetSuite, Gong and Veeva.

Writes run as the connecting user under that person's own permissions rather than through a shared service account, and other servers expose whatever operations their tools allow. Data arrives through Tellius Connect, which covers integration, preparation and the semantic layer, with live pushdown queries to cloud warehouses and up to 50 million rows in live mode on the Premium plan.

SourceTellius, Tellius 6.4 release post and tellius.com/pricing; tellius.com/resources/blog/tellius-6-4-put-kaiya-to-work-with-you-in-controlread 2026-10-05

A Mission starts from an outcome the team describes; Kaiya plans the steps, which users can shape before it builds, and orchestrates a squad of specialist agents across thousands of hypotheses into one brief. Missions are written as outcomes rather than scripted prompts, and they run either on their own or interactively with the user. In Plan mode, Kaiya reads the schema and writes the definition of an App or Mission before anything is built or run. A finished Mission can route its recommendation on to the next stakeholder, and Missions and Apps share one workspace.

SourceTellius, tellius.com/platform/ai-agents and the Tellius 6.4 release postread 2026-10-05

The customer's metrics, entity relationships and access rules are encoded in a governed semantic layer that Kaiya reasons against rather than raw SQL, so agents do not invent definitions or joins. Kaiya also arrives knowing industry terms such as NBRx and promo cycle and business structures such as territories, hierarchies and calendars. Connected sources include Snowflake, Salesforce, HubSpot, NetSuite, Gong and internal people and territory data.

SourceTellius, tellius.com/platform/ai-agents and /platform/appsread 2026-10-05

Every write is presented for review and executed only once approved, in conversation, in Apps and in Missions alike. The review shows the operation, the destination, the rows and the identity that will carry it out. Preview runs a Mission's plan as a draft so the team can check what each step returned, with a citation on every figure, before it is published. In Plan mode nothing is built or run while the user shapes the plan.

SourceTellius, tellius.com Tellius 6.4 release postread 2026-10-05

Tellius states SOC 2 Type II and HIPAA and supports SSO through SAML and role based access control. Data is encrypted in transit and at rest, Tellius says it does not train on user data, and row level policy is enforced at query time. Kaiya Apps inherit row level security, audit logs and SAML SSO from the platform, and SAML single sign on is listed as an Enterprise plan feature.

SourceTellius, tellius.com/platform, /platform/ai-agents, /platform/apps and /pricingread 2026-10-05

Missions show their execution step by step, including the SQL queries and Python analysis run, with full lineage and the statistics behind each finding so every number can be traced. Every figure in a Mission's summary carries a citation, published versions of Apps and Missions are kept with timestamps, and Apps include audit logs.

SourceTellius, Tellius 6.4 release post and tellius.com/platform/apps; tellius.com/platform/ai-agentsread 2026-10-05

Kaiya keeps memory that carries across sessions, including learned team vocabulary and confirmed queries that become defaults, and each Mission builds on the last; nothing is stored in the background, since a correction in chat offers a card the user chooses to save. The memory layers are named Skills, Phrase Learning and Query Learning. When saved memories shape an answer, they appear in Kaiya's thought process for that answer.

SourceTellius, tellius.com/platform/ai-agents and Tellius 6.4 release postread 2026-10-05

The Enterprise plan offers flexible deployment on the customer's cloud, on premises or on Tellius Cloud. The platform uses a single tenant architecture, with the data kept in the customer's own VPC. The Premium plan, for up to 10 users, is fully hosted on Tellius Cloud with 10 GB of data storage and five concurrent jobs, while Enterprise lifts those limits to unlimited users, data, jobs and feeds.

SourceTellius, tellius.com/pricing and /platformread 2026-10-05

A library of more than 14 named agents across pharma, CPG, FP&A and RevOps ships ready to configure, including NRx Drop Investigator, Territory Performance, Trade Promotion Optimizer, Variance Investigator, Revenue Anomaly Monitor and Pipeline Velocity Agent.

The rest of the list covers HCP Targeting Optimizer, Patient Adherence Tracker, Field Force Performance, Pricing Strategy Agent, Category Growth Detector, Churn Risk Detector, Executive Brief Agent and Compliance Watchdog. Beside the agents sit more than 50 prebuilt Kaiya Apps by function, such as Territory Performance Review, HCP Targeting Workbench, Brand Performance Review and Department Variance Monitor, each editable in plain English. Admins can publish App and Mission templates to their teams as Industry Packs.

SourceTellius, tellius.com/platform/ai-agents, /platform/apps and Tellius 6.4 release postread 2026-10-05

Missions run on a schedule and on events, alerts fire when a KPI breaks pattern, and briefings land in Gmail, Slack and Teams as finished documents. Monitoring runs around the clock, and the outputs include briefings, driver memos, PDFs and Slack summaries. Missions can also be reached through MCP, inside embedded applications, and from the browser with a keystroke inside Salesforce and Veeva.

SourceTellius, tellius.com/platform/ai-agentsread 2026-10-05

Admins choose the LLM Kaiya uses among OpenAI, Gemini and Azure OpenAI and supply their own keys and base URLs, a setting available from Tellius 5.4 onward. For Azure OpenAI, admins point Kaiya at their own resource endpoint and named deployment. There is no default model, so an admin picks the provider and model before Kaiya runs.

SourceTellius, help.tellius.com Setting up LLM for Kaiyaread 2026-10-05

Tellius runs as an MCP server so Claude, ChatGPT, Cursor or a team's own agent can call its governed definitions, and API access is listed on the Enterprise plan, but no tool list, authentication or endpoint reference is published. Enterprise pairs API access with embedding and white labeling. The Premium plan already includes model scoring through an API and embedded natural language search, Guided Insights and Vizpads.

SourceTellius, tellius.com/pricing and Tellius 6.4 release post; tellius.com/resources/blog/tellius-6-4-put-kaiya-to-work-with-you-in-controlread 2026-10-05

Preview runs a Mission's plan as a draft and shows what each step returned before it goes live; no test set, accuracy scoring or regression check is described. The draft shows the generated summary with a citation on every figure it reports. Earlier published versions of a Mission or App stay on record with their timestamps.

SourceTellius, tellius.com Tellius 6.4 release postread 2026-10-05

No browser or desktop operation by Kaiya is described; the browser appears among Mission surfaces only as a place where results arrive. Kaiya otherwise works through data connections, MCP servers and embedded applications.

SourceTellius, tellius.com/platform/ai-agentsread 2026-10-05