Day AI vs Rox (2026)

The agents write into outside systems as well as reading from them. A hubspot.sync tool pushes contact, company and deal data into HubSpot once a person approves, Salesforce, Gong, Gmail, Calendar, Slack and Zoom are named sources, and Zapier gives a route to systems Day AI has no connector for.

Sourceday.ai/productread 2026-10-01

Integrations cover Salesforce, with writeback, HubSpot, email, calendar, Slack, Microsoft Teams and data warehouses (Snowflake, Databricks, BigQuery), and the Teams plan includes CRM sync and writeback, so agents act in outside systems with the customer's credentials.

Sourcedocs.rox.com/development/product/integrations/salesforce/configuring-salesforce-integration-+-writeback.mdread 2026-10-01

Every agent can create and run workflows, and reusable skills such as an objection handling coach are deployed across many agents at once, so multi step work runs on a structure the customer configures, not one prompt at a time.

Sourceday.ai/productread 2026-10-01

Teams configure multi step, multi agent work. They build custom agents through several creation paths, attach triggers and outputs, and run them beside the prebuilt research, outreach, deal and meeting agents. Rox's agent framework orchestrates account aware swarms across research, outreach, opportunity management and proposal generation.

Sourcedocs.rox.com/development/product/custom-agents.mdread 2026-10-01

Work starts without a person asking. Workflows run on schedules, such as a daily 9:00 run, and on event triggers, and capture runs on meetings, email, Slack messages, calendar events and call completion as they happen.

Sourceday.ai/productread 2026-10-01

Custom agents wake on triggers without a person. They run on a schedule (every weekday at 8 AM), before or after a meeting, on transcript completion, on opportunity activity, on a webhook call from outside Rox, and when a Clever Column cell is generated. Tracked Clever Columns also re-run on a schedule and notify on change.

Sourcedocs.rox.com/development/product/custom-agents/agent-triggers.mdread 2026-10-01

The context graph is built over the customer's own record. It holds meeting recordings and transcripts, email threads, opportunities, enrichment and quotes with enterprise level permissioning, and it answers natural language questions across them, such as every demo the team ran this week.

Sourceday.ai/productread 2026-10-01

The corpus is the customer's own, and ingestion is described. A unified, governed knowledge graph consolidates structured sources including CRM, email and calendar data with unstructured ones including support tickets, job postings and news feeds, plus finance and product telemetry. The graph is built natively inside the customer's own warehouse, so entities, relationships and context across accounts, opportunities and interactions can be extracted reliably. Every agent action reasons over that graph, not over generic context.

Sourcemedium.com/snowflake/rox-the-next-generation-revenue-operating-system-powered-by-snowflake-08b5c6a9a963read 2026-09-01

Day AI names its memory layer, Customer Memory, as a product surface, and the company's premise is that the full conversational record persists instead of being compressed into CRM fields. What was said, promised and decided stays available across meetings, email and Slack, and the assistant reads it back to track commitments and prompt owners when timing matters.

Scope is the customer relationship and is addressable through natural language and through MCP tools that reach meeting transcripts and objects. Lifetime is governed by the published data policy covering retention and deletion. The context graph is what the agent retrieves to answer, while Customer Memory is what persists between engagements.

Sourceday.ai/product Customer Memory and day.ai/dataread 2026-09-01

There is no agent memory separate from the knowledge graph. Rox does not describe what an account agent retains from one cycle to the next or state a scope or lifetime for agent memory, and agents reason over the graph instead of writing to a memory they later read. The persistent understanding each account agent builds is that governed knowledge graph, the customer's own corpus.

Sourceaws.amazon.com/blogs/machine-learning/rox-accelerates-sales-productivity-with-ai-agents-powered-by-amazon-bedrockread 2026-09-01

A workspace setting requires a person's approval before the agent pushes data to an external CRM, shown in the product as the agent asking before it pushes to HubSpot, and outbound email is drafted for a person to send. That approval shows as product behavior, and no settings reference covers it.

Sourceday.ai/productread 2026-10-01

A person controls whether an agent runs and can judge what it produced. Custom agents run only once a team switches the Active toggle on, after a test run, and outputs carry approve and reject controls. No step holds a CRM write or email for that approval before it commits, so approval reviews the output and nothing gates the action itself.

Sourcedocs.rox.com/development/product/custom-agents/activate-monitor-and-review-outputs.mdread 2026-10-01

The trust center offers a SOC 2 Type II report to customers and prospects under NDA through a request form, and Day AI describes a maintained information security program.

Its named controls cover encryption of data at rest and in transit, access controls, a published subprocessor list, a data policy covering retention, location and deletion, and a signable data processing addendum, alongside terms and an acceptable use policy.

Access control for customers works through the MCP setup, where data access follows workspace role, capabilities follow the paid agent tier, and a customer can revoke a connected agent's tokens from workspace settings.

Sourceday.ai/trust and day.ai/mcpread 2026-09-01

Permission Sets bundle resource, field and custom resource permissions and are assigned to users or through Pods, organization wide defaults set record and field access, and Access Provenance traces who can view a record. Enterprise organizations set up SAML SSO with their own identity provider through a self serve wizard. Rox reports SOC 2 Type II compliance with annual independent audits, and its trust center lists a SOC 2 Type 2 report on request and CASA Tier 2, with ISO 27001 marked coming soon.

SourceRox docs governance and enterprise SSO pages, rox.com/security and trust.rox.com; docs.rox.com/development/product/governance/permission-sets.mdread 2026-09-30

Day AI says records created, updated and managed through its MCP server have all changes tracked and audited, and the product is built so a person can see what the agent wrote and override it. There is no run history, activity log view, decision trace or retention period for that audit, and the statement covers the MCP path, not the automatic capture that writes most of the record. That matters because the agent writing the system of record unattended is the product.

Sourceday.ai/mcp and day.ai/productread 2026-09-01

Agent runs are on the record. Each custom agent is monitored from its Agent page, a Runs page and the Home page, and its outputs (text, HTML, email, slides, files) can be reviewed, approved or rejected. Rox does not say what a run record shows, such as inputs, steps or tool calls, so why an agent acted cannot be reconstructed. Agentic monitoring tracks account signals, not agent runs.

Sourcedocs.rox.com/development/product/custom-agents/activate-monitor-and-review-outputs.mdread 2026-10-01

Day AI runs in the cloud only, with nothing the customer selects. There is no region choice, VPC, private tenancy, self-hosted or on-premises option. The data policy covers how long data is retained and how it is stored, handled and deleted, but not where customer data is hosted.

Sourceday.ai/trust, day.ai/data and day.ai/mcpread 2026-09-01

Execution inside the customer's data warehouse is the defining architectural property, and Rox says plainly that all customer data stays inside the customer's data warehouse.

Rox uses Snowflake zero-copy data sharing to gain governed access to warehouse data without persisting raw datasets outside customer control, with the Rox knowledge graph grounded natively in the customer's Snowflake instance.

That places the customer's data in their own environment and region by construction, a stronger residency position than naming regions and well beyond shared multi-tenant software as a service. The Rox application layer itself runs on AWS.

Sourceaws.amazon.com/marketplace/pp/prodview-hzr2usullmhoaread 2026-09-01

Each customer gets a single AI assistant it names for itself, and the CRM, meeting assistant and knowledge base arrive as one product, with no catalog, template gallery or named set of prebuilt agents. What it does offer is developer-shaped. The SDK ships three production-ready templates for desktop, mobile and serverless applications, and a second public repository, the Operations Harness, is offered for adoption.

Sourceday.ai/mcp SDK templates and day.ai/productread 2026-09-01

A named set of prebuilt agents each has its own documentation page and job. They are the Research, Outreach, Deal, Meet, Autofill and Apps agents, beside custom agents a team builds.

Sourcedocs.rox.com/llms.txtread 2026-10-01

Which models power the agents is not published, and the customer has no model choice.

Sourceday.ai/productread 2026-10-01

Rox's agents run on Amazon Bedrock, the infrastructure the platform is built on, not a choice offered to the customer, in the same way a cloud provider on a subprocessor list is infrastructure, not a second model. There is no model selector, per-agent choice, bring-your-own-key or routing policy, and no provider or model is offered for the customer to pick. Rox does not publish which models its agents use on Bedrock.

Sourceaws.amazon.com/blogs/machine-learning/rox-accelerates-sales-productivity-with-ai-agents-powered-by-amazon-bedrockread 2026-09-01

Day AI publishes an open SDK at github.com/day-ai/day-ai-sdk with an importable client, OAuth handling, a full MCP client and three production-ready templates covering Electron desktop, React Native mobile and Vercel serverless, and the site footer links it as API and SDK alongside a second public repository, the Operations Harness at github.com/day-ai/gtm-brain. Beside it runs an MCP server at day.ai/api/mcp over HTTP with OAuth 2.0, install paths for Claude, Cursor and Claude Code, a listing in the Claude connector directory, and more than twenty tools spanning object search, contact and opportunity creation, meeting transcript access, email drafting and notifications, with permissions inherited from agent tier and workspace role and tokens revocable from workspace settings.

Sourceday.ai/mcp and github.com/day-ai/day-ai-sdkread 2026-09-01

An outside system can start a Rox agent through the On webhook call trigger, which custom agents take for events that start outside Rox. There is no API reference, keys page, SDK or MCP server, so the webhook trigger is the only way in.

Sourcedocs.rox.com/development/product/custom-agents/agent-triggers.mdread 2026-10-01

The product has agents, workflows, skills, memory, MCP and an SDK, but no testing or evaluation tooling for agents.

Sourceday.ai/productread 2026-10-01

A customer runs a test of a custom agent before switching it on and reviews outputs on the Runs page afterwards, which is inspection, not a scored check. Rox's own developers run an Agent Evaluation System, with labeled datasets, accuracy metrics and LLM as a judge rubrics, on Rox's agents. There is no harness, scored test case or release gate a customer runs on its own agents.

Sourcedocs.rox.com/development/product/custom-agents/activate-monitor-and-review-outputs.mdread 2026-10-01

Browser or computer control is not part of the product, which covers agents, workflows, skills, memory, MCP and the SDK.

Sourceday.ai/productread 2026-10-01

Rox takes no action on an interface it does not control. Its agents create personalized emails and LinkedIn posts, which is generating content for a person to send or publish, not the agent operating LinkedIn. Everything else runs over programmatic routes, a knowledge graph built by warehouse sharing, connectors to CRM, support and finance systems, and delivery through web, Slack, macOS and iOS clients. There is no hosted or local browser, desktop session, remote computer control, scraping or third-party browser engine.

Sourcemarketplace.microsoft.com/en-us/product/rox-com.rox-enterprise-revenue-ai-agentsread 2026-09-01