Workflow Management
Tray.ai
Rest of world Report an error0–5 in half steps. 5 means the rubric's top anchor is met on the evidence.
by Tray.ai, Inc. · tray.ai
Report an error on this page Is this your product? →
Link to this page
Add ?theme=dark to the image address for dark backgrounds.
The badge shows the category and the month of the latest evaluation, and EU-Made where
that is the product’s provenance. It never shows a score.
Tray.ai at a glance
- Vendor
- Tray.ai, Inc.
- Headquarters
- Not stated on the vendor’s pages as of 13 Oct 2026
- Data hosting
- Default not stated; pages mention the US, the EU, United Kingdom and Australia source, captured 13 Oct 2026
- Free plan
- Not stated on the vendor’s pages as of 13 Oct 2026
- Panel verdict
- The AI panel rates Tray.ai 3/5, the mean of 6 judges. Tray.ai, Inc. how scores are made
- As of
- Vendor pages captured 1 Oct 2026 to 13 Oct 2026; evaluated by the panel 15 Oct 2026. all sources
Read this page as one judge. Each weighs the same scores by what they care about.
The panel's verdict
Tray.ai, Inc. of San Francisco offers an integration and automation platform on a SaaS basis, and the bench scored it strongest on integrations and connectors and on governance and audit, both at 7-8. A documented SAP Business One connector and custom connectors via the SDK are named, and an on-prem agent reaches internal services and databases over private IP without exposing public IPs. Governance rests on role-based access to per-connector and per-connection granularity, approval-gated promotion through dev, staging and prod, and decision records capturing who, when, what was shown and the outcome; retry behaviour and limits are published per error class. Hosting and portability scores cluster at 5, because the engine runs on the vendor's cloud and the on-prem agent is connectivity, not hosting, and sovereignty at 3, with a US contracting entity, an EU region on the Enterprise plan, and no public information on subprocessors. Process design scored 6-8; no criterion split the judges materially. The captured pages give different connector counts — 700+ and 474+ pre-built — and the pricing pages show no price figures.
Speaks for it
- Approvals route by role, with two approvers above higher-impact changes, deputy routing for an absent approver, and expiry defaulting to not acting.
- Dev, staging and prod are separated by approval gates, with audit logs of configurable retention streaming to Splunk or Datadog.
- The on-prem agent reaches internal services and databases over private IP without exposing public IPs.
- Projects and individual workflows export and import as JSON, including between organisations.
- Retry behaviour is documented per error class, alongside published payload, timeout and rate limits.
Held against it
- The platform is offered on a SaaS basis, and the on-prem agent is connectivity rather than a runtime the buyer controls.
- The EU region sits on the Enterprise plan and in the add-on list rather than as a standard choice.
- The vendor is recorded as Tray.ai, Inc. of San Francisco, and we found no public information on subprocessors or default execution location.
- The captured pages give different figures for the connector count — 700+ and 474+ pre-built.
- We found no public information on a public status page with incident history, failure alerting, SCIM provisioning or named certifications.
Best for
- You need agent-action approvals an auditor or works council can defend, with role-based routing and a decision record of who saw what.
- Your team must reach internal databases and services behind a firewall without exposing public IPs.
- You want one governance model across workflows, agents, MCP tools and pipelines, with audit logs streaming to your own SIEM.
- You automate SAP Business One and want connectors documented action by action, with templates.
Avoid if
- You need the engine to run on infrastructure you control — the service is offered on a SaaS basis and the on-prem agent is connectivity, not hosting.
- You require EU regional hosting as standard rather than an Enterprise-plan option.
- You are planning an exit to another platform — the documented export and import runs between the vendor's own orgs and workspaces.
The scores
| Criterion | Panel median |
|---|---|
| Process design, forms & approvals | 3.5 of 5 |
| Connectors, API & on-premise reach | 3.5 of 5 |
| Execution, error handling & monitoring | 3.5 of 5 |
| Governance, credentials & audit trail | 3.5 of 5 |
| Hosting options & portability | 2.5 of 5 |
| European sovereignty | 1.5 of 5 |
Process design, forms & approvals
Show reasoningHide reasoning
How this is scored
Modelling the process people take part in: a visual designer or BPMN, forms, approval steps with roles and substitutes, deadlines and escalation, and versioning of a process that is already running — judged on whether a business owner can change a process without breaking the instances in flight.
0 — Linear trigger-action chains only; no human steps, no forms and no statement on approvals or roles.
3 — A visual builder with branching and a simple approval step, but approvers are named individuals rather than roles, there are no deadlines or escalation, and changing a process has no versioning.
5 — A visual designer with conditions and parallel paths, built-in forms, approval steps assigned to roles or groups, deadlines with reminders, and versions of a process kept with the running instances on their original version.
8 — BPMN 2.0 modelling or an equivalent documented notation, substitutes and escalation chains, multi-stage approvals with four-eyes rules, forms in German with validation and file upload, a task inbox for participants, and a test or simulation mode before publishing.
10 — The process is a governed asset: version history with diff and rollback, migration of running instances to a new version documented, process analytics on cycle time and bottlenecks per step, and the design documented well enough that a business team can maintain it without the vendor.
The Process Owner
Approvals resolve from group ownership rather than named individuals, an absent approver routes to a deputy, a request that expires defaults to not acting, and higher-amount cases go to two approvers — the four things I insist on for purchase approvals. Project versions snapshot workflows and configuration with change lists, named authors and restore, an observe mode classifies gating rules without blocking before go-live, and a form trigger accepts file attachments. I found no public information on a documented process notation such as BPMN, on a task inbox for participants, or on what happens to requests already in flight when a new version is published. 5 6 2 8
The Integration Engineer
Agent action approvals are built like real approvals: requests route by role — account owner for customer-facing, finance for money, system owner for data, two approvers above a higher blast-radius bar — an absent approver falls back to a deputy, an unanswered request expires into not acting, and every decision is recorded with who, when, what was shown, the outcome and the aftermath, with an observe mode to dry-run gating before production. Project versioning carries change lists with authors, restore, and rollbacks that archive rather than delete stored data. I found no public information on BPMN or an equivalent documented notation, on migrating running instances to a new version, or on per-step cycle-time analytics. 5 6 2
The Operations Analyst
Approval handling is unusually well thought through: requests route by role — account owner for customer-facing, finance for money, system owner for data — with two approvers above a higher cut-off, a deputy for an approver who is away, and expiry that defaults to not acting. Project versions keep a change list with authors and can be restored or rolled back without destroying stored data, and approvals report median time to decision. We found no public information on a documented process notation such as BPMN, a task inbox for participants, or how in-flight instances behave when a new version is published. 5 6 2 8
The Head of IT
Approval design is genuinely role-based — customer-facing work to the account owner, money to finance, data changes to the system owner, two approvers above a higher impact level, deputy routing when an approver is away, and expiry that defaults to no action — and project versioning keeps a change list with authors plus restore and rollback that archives data rather than deleting it. Definitions are JSON and we found no public information on BPMN or an equivalent documented notation, a task inbox for participants, or what happens to instances in flight when a new version publishes. 5 6 13
The Data Protection Officer
Approvals are designed as transfers I could defend to a works council: routing by owning role, two approvers above a higher blast-radius limit, deputy routing when an approver is away, expiry that defaults to not acting, and a decision record of who, when, exactly what was shown, the outcome and what happened after execution. Projects are versioned with author tracking, change lists and a rollback that archives rather than deletes, and observe mode plus per-step schema checks exist before anything touches production; I found no public information on BPMN-style notation, a participant task inbox, or what happens to instances already running when a new version is published. 5 6 1 9
The Skeptic
Approvals are the strongest documented part: requests routed by role — account owner for customer-facing, budget owner for finance, system owner for data — with two approvers above higher-impact changes, deputy routing when an approver is away, expiry that defaults to not acting, and an observe mode before anything is gated. Project versioning records who changed what, with restore and a rollback that archives rather than deletes stored data. We found no public information on a modelling notation such as BPMN, a participant task inbox, forms in German with validation, or how instances already in flight behave when a new version publishes. 5 6 13
Connectors, API & on-premise reach
Show reasoningHide reasoning
How this is scored
Reaching the systems a European company actually runs: maintained connectors judged by name rather than count, a generic HTTP and webhook step, custom connectors, and a way to reach systems behind the firewall without opening it.
0 — No connectors beyond email, and no API or webhook step.
3 — A large connector count on a logo wall, with no statement of which actions each supports or who maintains them, and no generic HTTP step.
5 — Named connectors with their triggers and actions documented, including Microsoft 365 and at least one ERP or accounting system common in DACH (e.g. SAP, DATEV, Microsoft Dynamics), a generic HTTP/REST step, inbound webhooks, and a documented API to start and query workflows.
8 — Custom connectors buildable from an OpenAPI description or code, an on-premise agent or self-hosted runner reaching internal databases and file shares, SOAP and database steps, OAuth handled per connection, and connector maintenance ownership (vendor, partner or community) stated.
10 — Integration is engineered: a versioned public API with a deprecation policy, connector changelogs, rate limits per connector published, event-driven triggers rather than polling where the source allows it, and the vendor maintaining the German-market connectors (DATEV, SAP, ELO or d.velop) itself.
The Process Owner
Named connectors come with their read and write direction documented, the SAP Business One connector lists its business actions individually, and a universal HTTP client, inbound webhook triggers and documented API endpoints are all published. The on-prem agent reaches internal services and databases over private IP addresses without opening them to the internet, and custom connectors can be built with the connector SDK. I found no public information on who maintains each connector, on SOAP steps, or on API versioning and deprecation. 5 7 12 14 8
The Integration Engineer
The SAP Business One connector is documented by action rather than logo — Create Sales Orders, Create and Post Invoices, Log Journal Entries, Manage Service Calls against named data sources — with Microsoft Dynamics 365 named as an alternative stack, a generic HTTP Client, inbound webhooks, custom connectors via the SDK, and an on-prem agent that talks to internal services and databases over private IPs without opening anything to the internet. The captured pages give different figures for the connector count (700+ on one page, 474+ pre-built on another) and are silent on who maintains each connector. I found no public information on DATEV, SOAP steps, connector changelogs or an API deprecation policy. 7 12 14 8
The Operations Analyst
Named connectors come with their actions documented — the SAP Business One pages list agent tools and data sources, Microsoft Dynamics 365 and Teams appear by name, and the generic HTTP client, webhook trigger and public API carry published timeouts and limits. Custom connectors are buildable with the SDK, and an on-prem agent reaches internal services and databases without exposing them to the internet, alongside VPN, PrivateLink and IP allow-listing options. The captured pages give different figures for the connector count (700+ and 474+ pre-built), and we found no public information on connector maintenance ownership, a deprecation policy, or per-connector rate limits. 5 7 11 12 14 8
The Head of IT
The on-prem agent tunnels to internal services and databases without opening them to the internet, with VPN and AWS PrivateLink options documented alongside — exactly the reach we need behind our firewall. Connectors are named with their read and write actions (Okta, Salesforce, Slack, Jira, Snowflake), SAP Business One and Microsoft Dynamics 365 appear, custom connectors are buildable via the connector SDK, and a universal HTTP client plus webhook triggers round it out. We found no public information on who maintains each connector, or on published per-connector rate limits and an API deprecation policy. 5 7 12 14 8
The Data Protection Officer
The captured pages give different figures for the connector count (700+ versus 474+ pre-built), but the catalogue names connectors with an on-premise agent that reaches internal databases over private IP or proxy without opening the firewall, a generic HTTP client with a documented 15-minute timeout, inbound webhooks with published limits, custom connectors via the SDK, and a documented SAP Business One offering for the DACH estate. I found no public information on who maintains each connector, a DATEV-class German connector, or a versioned API with a deprecation policy and per-connector rate limits. 14 12 7 8 11
The Skeptic
A generic HTTP client, inbound webhooks with a published 429 rate limit, custom connectors built with a connector SDK, an on-prem agent reaching internal services and databases without exposing public IPs, and a documented SAP Business One connector with templates make the reach concrete rather than a logo wall. The captured pages give different figures for the connector count — 700+ on the homepage and pricing pages, 474+ pre-built on the connectors index. We found no public information on who maintains each connector, per-connector documentation of triggers and actions, SOAP steps, or DATEV. 7 12 14 8
Execution, error handling & monitoring
Show reasoningHide reasoning
How this is scored
What happens when a step fails at 3 a.m.: retries, error paths, alerting, an execution history someone can search, limits on runtime and throughput, and a status page that admits incidents.
0 — Nothing stated about failed runs; an error stops the workflow silently.
3 — Failed runs are marked and an email is sent, but retries are manual, execution history is short or undocumented, and runtime and throughput limits are unstated.
5 — Automatic retries with configurable back-off, an error path or error workflow, a searchable execution log with input and output per step, failure alerts by email or chat, and runtime, payload and rate limits documented.
8 — Replay of a failed run from the failing step, idempotency guidance, queueing and concurrency controls, execution history retention stated per plan, monitoring exportable to the customer's own tools, and a public status page with incident history and an uptime commitment.
10 — Operations are designed for production: an SLA with service credits, horizontal scaling or high-availability setup documented for self-hosted installs, dead-letter handling, end-to-end tracing across workflows, and post-incident reports published.
The Process Owner
Retries are documented per error class with escalating delays, runs can be re-run from the logs within a seven-day window, log retention is stated per plan, and payload, timeout and rate limits are published down to the megabyte. Runs, agent actions and tool calls are logged as searchable detail and can be streamed to our own SIEM such as Splunk or Datadog. I found no public information on a public status page with incident history, on service-level commitments with credits, or on queueing and concurrency controls. 9 10 8
The Integration Engineer
The retry and limits documentation is unusually candid: backend errors retry three times, internal errors fifteen with growing delay, wrong user input never retries, connector timeouts run 45 seconds (up to 60 minutes for Redshift), payload caps and webhook rate limits are published, log retention is stated per package, and logs can be re-run for 7 days. The captures do not say a failed run replays from the failing step, and the 100 percent execution uptime figure appears as a marketing claim rather than an incident ledger. I found no public information on a public status page with incident history, SLA service credits, queueing and concurrency controls, or dead-letter handling. 8 10 2
The Operations Analyst
The technical-limits page is exactly what I ask for before a month-end run: payload caps between steps, connector and webhook timeouts, and retry behaviour per error class, with internal errors retried fifteen times on rising delays. Logs are searchable, re-runnable for seven days, retained seven or thirty days by package, and streamable to Splunk, Datadog or our own SIEM. We found no public information on failure alerts reaching email or a chat channel, idempotency guidance, or a public status page with incident history — only a claim of "100% execution uptime". 8 9 10 2 8
The Head of IT
Retry behaviour is documented by error class — fifteen retries with growing delay on platform errors, three on connector errors, none on user-input mistakes — and payload, timeout and throughput limits are published per operation, with log retention of 7 or 30 days by package, a seven-day re-run window, and log streaming plus SIEM export available. Enhanced SLAs are mentioned for the top plans, but we found no public information on their terms or service credits, on failure alerting by email or chat, or on a public status page with incident history. 8 9 10 2 8
The Data Protection Officer
Failure behaviour is documented in unusual detail: three retries for connector errors, fifteen escalating retries for internal errors, an outcome field that separates third-party failure from a 200 response, and logs that can be re-run for seven days. Retention is stated per package (7 or 30 days, reducible on request), payload and rate limits are published as concrete numbers, and audit and run logs stream to the customer's own SIEM; I found no public information on a status page with incident history, a service-credit SLA, or queueing and concurrency controls. 8 8 9 10 2
The Skeptic
Retry behaviour is documented to an unusual depth — connector failures retry 3 times, internal backend errors 15 times with growing delay, wrong user input never retries — and payload, timeout and rate limits are public, with log retention of 7 or 30 days by package and logs re-runnable for 7 days. What we did not find is what wakes a human: we found no public information on error paths, failure alerting by email or chat, or a public status page with incident history. "100% execution uptime" appears as a marketing claim, and the content of the "enhanced SLAs" is not published. 8 9 10 8
Governance, credentials & audit trail
Show reasoningHide reasoning
How this is scored
Keeping automation from becoming shadow IT: roles and permissions, SSO, where connection credentials are stored, separate environments with controlled publishing, an audit trail of changes and approvals, and what the execution log retains of the payload. An approval record is also a record of employees' behaviour, so what the trail captures matters to a works council.
0 — Single-user accounts or one shared login; no roles, no audit log and no statement on credential storage.
3 — Admin and member roles and shared team folders, but credentials are visible to anyone who can edit a workflow, there is no audit log, and execution payloads are kept without a stated period.
5 — Role-based permissions per workflow or folder, SSO via SAML or OIDC, credentials stored encrypted and usable without being revealed, an audit log of workflow changes, and the retention period of execution data stated.
8 — Separate development, test and production environments with a publishing approval, SCIM provisioning, external secret stores supported, payload redaction or log-off per workflow, configurable execution data retention, and the audit trail exportable — with the fields it records about users described.
10 — Governance a works council and an auditor can both sign off: field-level masking of personal data in logs, delegated administration per department, a central inventory of every workflow and connection with its owner, certifications published (ISO 27001 or SOC 2), and documentation of what the platform records about individual users.
The Process Owner
Access control is role-based per workspace, connector, connection and tool; credentials live as workspace authentications referenced by name so the workflow never holds them; and development, staging and production are separated by approval gates. The approval record captures who, when, exactly what was shown, the outcome and what happened after execution — what a works council will ask to see — and project change history names the author of every edit, with log retention configurable and streamable. I found no public information on SCIM provisioning, external secret stores, or the compliance certifications themselves. 5 6 10 8
The Integration Engineer
Permissions run per workspace, per connector, per connection and per MCP tool; dev, staging and production are separated by approval gates with sign-off before anything reaches production; credentials live as named workspace authentications the workflow never holds; and audit logs are searchable, retention-configurable and streamable to Splunk or Datadog, with masking for PII and financial data and SSO available. Decision records capture who, when, what was shown, the outcome and what happened after execution — the part a works council will actually ask about. I found no public information on SCIM provisioning, external secret stores, or named certifications; a trust page is referenced but its contents were not captured. 10 5 2 3
The Operations Analyst
Role-based access reaches per workspace, per connector, per connection and per MCP tool, dev, staging and prod are separated by approval gates with sign-off before production, and audit logging is configurable in retention and streams to our SIEM with every action traced to a real user. Execution data retention is stated at seven or thirty days by package, log masking is available, credentials are workspace authentications referenced by name so workflows never hold them, and version history records who changed what. We found no public information on SCIM provisioning, external secret stores, a statement that stored credentials are encrypted, or a named certification such as ISO 27001 or SOC 2. 2 5 6 10 8
The Head of IT
This is close to what I insist on: dev, staging and prod separated by approval gates with sign-off before production, role-based access down to per-connector and per-connection granularity, audit logs with configurable retention streaming to Splunk or Datadog, built-in masking of personal and financial data, and a named owner on everything shipped. Credentials live as workspace authentications referenced by name so workflows never hold them. We found no public information on SCIM provisioning, external secret stores, or named certifications such as ISO 27001 or SOC 2, and SSO sits in the paid add-on list. 1 2 5 10 2
The Data Protection Officer
This is the part I would actually sign: role-based access down to workspace, connector, connection and MCP tool, approval-gated promotion from dev through staging to production, execution log retention stated in days and reducible on request, masking of PII, PHI and financial data in workflows, and an audit trail that streams to Splunk or Datadog. The approval record captures who, when, exactly what the approver was shown and the aftermath — precisely what a works council will ask for; I found no public information on SCIM provisioning, external secret stores, or a published certification such as ISO 27001 or SOC 2, only a pointer to a trust page. 10 5 6 8 2 4
The Skeptic
One role model covers workflows, agents, MCP tools, APIs and pipelines at per-connector and per-connection granularity, dev/staging/prod sit behind approval gates, every change is recorded with its author, and data masking plus configurable log retention with SIEM streaming cover the payload side. Credentials are workspace authentications referenced by name, so a workflow never holds them. We found no public information on SCIM provisioning, external secret stores, encryption of the credential store, or named certifications — the security page is referenced but the certifications it holds are not named in what we captured. 5 6 10 2
Hosting options & portability
Show reasoningHide reasoning
How this is scored
Where the engine runs and whether the workflows can leave: cloud regions, self-hosted or on-premise editions, the licence of the engine, and export of workflow definitions in a format another tool or a later version can read.
0 — Cloud-only with no stated region, and no export of workflow definitions.
3 — Cloud-only with a region named, and an export that produces a proprietary file only the same product can import.
5 — An EU cloud region selectable, workflow definitions exportable as documented JSON or BPMN 2.0, and an API to read definitions and execution history for backup.
8 — A self-hosted or on-premise edition with its system requirements documented, workflow definitions versionable in Git, the engine's licence published, and a documented migration path between cloud and self-hosted in both directions.
10 — The customer owns its automation: an open-source or source-available engine, definitions in an open standard executable elsewhere, air-gapped installation supported, and a stated exit process that hands back definitions, credentials metadata and execution history in documented formats.
The Process Owner
Hosting regions include the EU, but regional hosting sits on the Enterprise plan or as a paid add-on, and the on-prem offering is a connectivity agent rather than an engine we could run ourselves. Projects and individual workflows export and import as JSON files, including between organisations, and version data exports as JSON too. I found no public information on a self-hosted edition with system requirements, on Git versioning of definitions, or on the engine's licence. 1 10 12 13
The Integration Engineer
The engine runs on Tray's managed SaaS runtime, and the on-prem story is an agent plus tunnel options for reaching internal systems — I found no public information on a self-hosted edition of the engine itself, on an engine licence, or on a migration path between cloud and self-hosted. What is evidenced: regional hosting in US, EU and APAC with the EU region as a paid add-on on the Enterprise plan, project and workflow export and import as JSON, and deployment steps automatable through the Projects and Solutions APIs. 13 2 12 10
The Operations Analyst
Workflows run on the vendor's managed runtime with US, EU and APAC regions, though region-specific hosting sits on the Enterprise plan and appears as a listed add-on. Projects and workflows export as JSON with documented import and overwrite modes between orgs and workspaces, and version history can be exported as JSON for backup. We found no public information on a self-hosted or on-premise edition of the engine itself, its licence, or an air-gapped installation — the on-prem agent is connectivity to our systems, not a runtime we control. 1 10 12 13 2
The Head of IT
Projects and individual workflows export as documented JSON including configurations, data tables and agent metadata, with API endpoints automating the deployment steps — a workable artefact for backup. But the engine is offered on a SaaS basis: the on-prem agent reaches inward to internal systems while execution stays on the vendor's cloud, the EU region is selectable only on the Enterprise plan and as a listed add-on, and we found no public information on a self-hosted edition, an engine licence, or a migration path off the cloud. That misses the exit route I require before a department commits. 1 4 12 13 2
The Data Protection Officer
The engine is cloud-hosted with a selectable EU region (US, EU and APAC, with region-specific hosting reserved for the Enterprise plan and as a priced add-on name), and projects and individual workflows export as JSON that the same platform re-imports, between orgs and workspaces. The on-premise agent, tunneling and PrivateLink options reach internal systems but that is connectivity, not hosting of the engine; I found no public information on a self-hosted edition, the engine's licence, or an exit path into a format another tool could execute. 10 13 12 1 2
The Skeptic
Projects and individual workflows export as documented JSON — workflows, project configuration, solutions, data tables and more — importable between orgs, with historical versions exportable the same way and a CLI positioned for SDLC use. US, EU and APAC hosting is documented, but regional hosting sits on the Enterprise plan and in the add-on list, so an EU execution location is a purchasable option rather than a default. We found no public information on a self-hosted or on-premises edition of the engine itself or its licence; the on-prem agent is a connection to internal systems, not a place the engine runs. 1 6 10 13
European sovereignty
panel opinion
Show reasoningHide reasoning
How this is scored
Who the contracting entity is, where workflows execute and where execution logs with their payload copies are stored, and who the subprocessors are — including the AI model providers behind any AI steps. Independently sourced by the sovereignty pipeline. A self-hosted edition counts only for the data that actually stays on the customer's servers.
0 — Non-EU vendor and contracting entity, execution location unstated, and subprocessors unnamed.
3 — An EU region offered as an option while the contracting entity is non-EU, or the subprocessor list is absent, or AI steps send data to model providers of unstated location.
5 — EU execution and log storage as standard and an EU contracting entity, but parts of the chain — AI steps, support access, telemetry — are non-EU without an explained safeguard.
8 — EU execution on named infrastructure, EU contracting entity, subprocessor list published with locations including AI providers, a DPA covering execution payloads, and a self-hosted option that sends no workflow data to the vendor.
10 — Sovereign end to end and evidenced: European ownership, entity, hosting and every subprocessor, AI steps runnable on European or self-hosted models, certification published, and a self-hosted edition that works without any call home.
The Process Owner
The contracting entity is the US company and the EU is one of three selectable hosting regions, offered on the Enterprise plan; I found no public information on the subprocessor list or on where AI steps process data, with US model providers such as OpenAI, Anthropic and AWS Bedrock among the LLM connectors. The only AI safeguard stated is that customer data is not used to train models. For a German mid-sized company I found no public information on an EU contracting entity, EU execution as standard, or a data-processing agreement covering execution payloads. 1 4 10 14
The Integration Engineer
An EU region exists, but as a purchasable add-on on the Enterprise plan while the vendor is US-based Tray.ai, Inc. of San Francisco, and the captured pages confirm nothing about the default execution location or the contracting entity. The subprocessor list is absent from the captures, and AI steps draw on model providers such as Anthropic, AWS Bedrock and OpenAI with no stated location for the data those steps send. 1 2 14 10
The Operations Analyst
The vendor is US-based (San Francisco) with a UK subsidiary, and an EU region exists only as an option — an Enterprise feature and a listed add-on — rather than as the standard. We found no public information on the subprocessor list, including which model providers sit behind the AI steps, on a data-processing agreement covering execution payloads, or on contracting-entity jurisdiction confirmed on the vendor's own captured pages. For a European buyer this reads as a US platform with an EU hosting choice. 1 10 2
The Head of IT
The contracting entity is a San Francisco company, an EU region is offered as an option rather than the default, and the LLM connector catalogue names US model providers. We found no public information on a subprocessor list, on the default location of execution and log storage, or on where the AI model providers behind Merlin AI process data; the commitment not to train models on customer data is a safeguard, but it is not residency. 4 10 2 14
The Data Protection Officer
The vendor is presented as a San Francisco company, the privacy terms read as an American notice (45-day response windows, recorded sales calls, an opt-out measured in business days), and the captured pages do not confirm an EU contracting entity, so an EU hosting region on the Enterprise plan does not make the chain European. I found no public information on where executions and their payload copies are stored by default, on a subprocessor list, or on the locations of the AI model providers behind the Merlin steps, even though the connector catalogue names OpenAI, Anthropic, Google and others; the stated no-training commitment on customer data is welcome but says nothing about where the AI steps run. 3 4 10 14 1
The Skeptic
The vendor is recorded as Tray.ai, Inc. of San Francisco, USA, with an EU region available only as Enterprise-plan regional hosting rather than a standard, and the vendor's own captured pages confirm neither the contracting entity's jurisdiction nor where execution and logs land by default. We found no public information on subprocessors, so where anything downstream runs is unstated. The AI connectors reach model providers including OpenAI, Anthropic, Gemini, Grok and DeepSeek with no stated location or safeguard for what those steps send. 1 10 14
Pricing transparency
not rated — the vendor publishes no price
Show reasoningHide reasoning
How this is scored
Whether a buyer can compute the monthly cost of their own processes from public pages alone: what the billing unit is (task, operation, execution, run, user or flow) and what counts as one, the allowance per tier, overage prices, and the cost of premium connectors, self-hosting, environments and AI steps.
0 — No public prices at all; every tier is a sales conversation.
3 — A starting price exists, but the billing unit is undefined or what counts as one task or execution is unstated — the invoice is unknowable.
5 — Tier prices public with the billing unit defined and the monthly allowance given, but at least one commonly needed piece (overage, premium connectors, extra environments, self-hosted licence, AI steps) is unpriced.
8 — Every tier priced publicly with its unit, allowance and limits, overage priced per unit, premium connector and add-on prices listed, self-hosted licensing stated, minimum term and VAT treatment given.
10 — Complete price computability: the monthly cost derivable for a given number of workflows, runs and steps per run, with a published calculator or worked examples, every add-on and overage on a public page, and the effect of loops and retries on the count explained.
The Process Owner
The pricing pages name the tiers, their workspace counts and insight windows, and list the add-ons by name — SSO, regional hosting, HIPAA, log retention, log streaming — but I found no public prices on any of them and no definition of the billing unit or of what counts as one execution. From what is published I cannot compute what our approvals would cost per month; every tier is a sales conversation. 2 2
The Integration Engineer
The captured pricing pages name the Team and Enterprise plans with their workspaces, insight windows and a long add-on list — HIPAA, SSO, regional hosting, log retention, log streaming, data engineering — but show no price figure of any kind: no tier price, no billing unit, no allowance, no overage rate. A free trial and an upgrade-anytime statement are all the pages offer. I found no public information on the price of any add-on, on AI steps, or on how runs or tasks are counted. 2 2
The Operations Analyst
The pricing pages name the tiers and the add-on catalogue — HIPAA, SSO, regional hosting, log retention, log streaming — with workspace counts and retention days, and a free trial, but we found no public information on any price, a starting figure, or a billing unit defining what counts as a task, execution or run. I cannot begin to compute what a loop over five thousand rows would cost; beyond the trial, buying this reads as a sales conversation. 1 2 2
The Head of IT
The pricing pages name a Team and an Enterprise plan with feature lists and a long add-on catalogue — SSO, regional hosting, log retention, log streaming — but we found no public information on any price for a tier or add-on, on the billing unit, or on a monthly allowance. Downgrades go through the support team, and a buyer cannot compute a monthly cost from the public pages. 2 2
The Data Protection Officer
The pricing pages name plans and enumerate add-ons by name — HIPAA, SSO, regional hosting, 30-day log retention, 180-day insights, log streaming — but state no figure for any tier or any add-on, and I found no public information on the billing unit, what counts as one unit, the allowance per tier, or overage pricing. A buyer cannot compute a monthly cost from what is public, and plan downgrades are routed to a support conversation. 2 2
The Skeptic
The pricing pages name a Team and an Enterprise plan with feature lists and named add-ons — SSO, regional hosting, 30-day log retention, 180-day insights visibility, log streaming — plus a free trial, but no price figure appears on the captured pricing pages. We found no public information on the billing unit, what counts as one task or execution, monthly allowances in those terms, overage, or the price of any add-on. The monthly cost of a buyer's own processes is not computable from what is public. 1 2 2
European sovereignty — proven facts
0 of 4 dimensions provenBuilt only from facts shown on the vendor's own pages. A dimension we could not prove is left open, not scored as zero.
| Legal entity | Not determined ⚠ unverified | — | uncited Report an error |
|---|---|---|---|
| Ownership | Not determined | — | uncited Report an error |
| Data residency | Not determined ⚠ unverified | — | uncited Report an error |
| Subprocessors | Not determined | — | uncited Report an error |
Questions about Tray.ai
Is Tray.ai European?
No. whats-best.ai classifies Tray.ai as Rest of world.
Where does Tray.ai store data?
The default storage location for customer data is not stated on the vendor’s pages as of 13 Oct 2026. The vendor’s pages mention the US, the EU, United Kingdom and Australia as hosting locations, per the vendor’s page captured 13 Oct 2026.
Does Tray.ai have a free plan?
Not stated on the vendor’s pages as of 13 Oct 2026.
What are the best alternatives to Tray.ai?
As of 15 Oct 2026, the panel rates n8n (3/5), Camunda (2.5/5) and Frends (2.5/5) highest among the other 16 Workflow Management products evaluated. Ratings that look equal are rounded to half stars; the order uses the unrounded rating, then the name. The highest-rated European alternative is n8n (EU-Made). Ratings are the AI panel’s opinion, formed from public evidence: how scores are made. The full list, with European products first: Tray.ai alternatives.
Where this could be wrong
- Evidence ages. The oldest capture behind this page is from 1 Oct 2026. Vendors change pricing and policies without notice; every fact reflects its source as of the capture date shown in the registry.
- Weak sourcing — Legal entity, Data residency. Not confirmed on the vendor’s own pages as captured.
- AI can misread a source. Extraction and judgement are automated; a citation guarantees traceability, not infallibility. If something here is wrong, say so — no account needed, every report is decided within 5 business days, and accepted corrections are published.
What we left out
A claim that does not survive our checks costs us the claim, not the page. This is what was taken off this one.
- 32 product facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 8 integrations facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 6 legal facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 3 pricing facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 data facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 subprocessors facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 2 sovereignty dimensions could not be confirmed on the vendor’s own pages and are shown as unknown. Know more? Tell us
- 6 of the readings below were written against an earlier fact sheet — a fact has been corrected, added or pulled since. Until the panel next runs on this product you are reading the older judgement. Know more? Tell us
Sources (14)
The pages every claim on this page was read from — each one checked, dated, and kept verifiable.
- 1 Vendor homepage tray.ai Checked 13 Oct 2026 Details →
- 2 Pricing page tray.ai Checked 13 Oct 2026 +1 earlier capture: 1 Oct 2026 Details →
- 3 Privacy policy tray.ai Checked 13 Oct 2026 Details →
- 4 Terms of service tray.ai Checked 13 Oct 2026 Details →
- 5 Process design, forms & approvals — found from sitemap tray.ai Checked 13 Oct 2026 Details →
- 6 Process design, forms & approvals — found from sitemap tray.ai Checked 13 Oct 2026 Details →
- 7 Connectors, API & on-premise reach — found from sitemap tray.ai Checked 13 Oct 2026 Details →
- 8 Execution, error handling & monitoring — found from sitemap tray.ai Checked 13 Oct 2026 +1 earlier capture: 1 Oct 2026 Details →
- 9 Execution, error handling & monitoring — found from sitemap tray.ai Checked 13 Oct 2026 Details →
- 10 Governance, credentials & audit trail — found from sitemap tray.ai Checked 13 Oct 2026 Details →
- 11 Governance, credentials & audit trail — found from sitemap tray.ai Checked 13 Oct 2026 Details →
- 12 Hosting options & portability — found from sitemap tray.ai Checked 13 Oct 2026 Details →
- 13 Hosting options & portability — found from sitemap tray.ai Checked 1 Oct 2026 Details →
- 14 Connectors, API & on-premise reach — found from sitemap tray.ai Checked 13 Oct 2026 Details →