Subprocessor Schedule
This schedule identifies providers Task Machine may use to process customer personal data and separately identifies important independent controllers. Provider names identify the service and legal entity where verified. A provider receives data only when the relevant feature, configuration, or instruction uses it.
Last updated:
Core subprocessors
| Provider | Purpose | Data | Location and transfer notes |
|---|---|---|---|
| Cloudflare, Inc. | Authoritative DNS, application reverse proxy, web application firewall protection, TLS termination, and Turnstile abuse prevention. | DNS queries and resolver metadata. Client IP addresses, URLs, request and response headers, cookies, authentication material, and customer content passing through the proxy. Turnstile challenge tokens and browser, TLS, and interaction signals. | Global network, including processing outside the EEA. Cloudflare terminates proxied TLS and uses a separate encrypted connection to our Scaleway origin. Its self-service agreement incorporates its DPA and international-transfer provisions. Primary application storage remains in Scaleway. Cloudflare caches and security records have separate lifecycles. |
| Scaleway | Application hosting, managed PostgreSQL, object storage, networking, static and CLI delivery with supporting DNS, transactional email, observability, and backups. | Account, workspace, workflow, transcript, upload, email, log, metric, and backup data. | Task Machine configures its primary application infrastructure and managed services in the EU. |
| Boat | Managed agent sandboxes, processes, and checkpoint-backed working files. | Task and chat source files, repository history, working files, runtime requests, process state, and provider resource metadata. | Boat states that its sandboxes run on machines in Germany, Finland, and France. Task Machine cannot choose a datacenter. This does not mean all related provider processing, network routing, logs, or backups occur only in the EU. |
| Blaxel | Managed agent sandboxes, processes, and checkpoint-backed working files. | Task and chat source files, repository history, working files, runtime requests, process state, and provider resource metadata. | Task Machine requests Frankfurt, Germany, as the infrastructure region for managed cloud sandboxes. This does not mean all related provider processing, network routing, logs, or backups occur only in that region. |
| DataFast | Cookieless public website acquisition analytics, consent-based persistent cross-domain journey measurement, AI crawler reporting, and revenue attribution. | Public page paths, referrers, campaigns, IP address, user agent, pseudonymous visitor and session identifiers, crawler requests, and linked payment attribution. | DataFast-managed service. Cookieless browser analytics runs before acceptance and after rejection. Persistent identifiers, cross-domain linking, and checkout attribution require acceptance. Contractual processing location and transfer details apply under DataFast's service terms. |
| OpenRouter | Gateway and routing for Task Machine managed model requests, including structured evaluations and optional voice dictation. | Prompts, context, messages, tool definitions and calls, generated responses, bounded evaluation state, typed questions and answers, model identifiers, token usage, request metadata, and workspace or run correlation needed for routing and settlement. Optional voice input also processes short audio phrases, authorized vocabulary hints, and transcription text. | Global service. Processing locations depend on the selected model and provider arrangements. |
| Composio (Sampark, Inc.) | Managed account authorization, token custody and refresh, and tool execution for supported Connectors. Connections using your own OAuth app go directly to the service without Composio. | Workspace-scoped identifiers, selected integrations and connected accounts, OAuth credentials, tool arguments and results, execution metadata, and temporary tool files. Upstream tokens are held by Composio rather than supplied to Workers. | Task Machine requires Composio's data processing agreement and payload zero data retention for managed Connector calls. The payload requirement covers tool-call and Proxy Execute payloads, not credentials, audit logs, operational metadata, temporary files, or existing product history. Managed Connector processing is separate from Task Machine's managed-AI zero-data-retention setting. Composio describes its terms in its Data Processing Addendum. Read its zero-data-retention policy and Composio's data-retention policy. |
Task Machine automatically stores private checkpoints of Local Worker and managed cloud sandbox working directories in Scaleway Object Storage in Amsterdam, Netherlands. Gitignored files are excluded when the directory has a remotely reachable Git base. Other directories are copied in full, including hidden files and credentials present there. Scratch files are not included.
Private working-file checkpoints are the source of truth for Local Workers and managed cloud sandboxes. Sandbox disks are temporary caches. Task working files expire 14 days after the last update, or 7 days after completion, cancellation or archiving, whichever comes first. Chat working files expire 14 days after last activity or 7 days after archiving, whichever comes first. Other work scopes expire after 14 days without activity. Repeating an archive request does not restart the clock. Restoring a Task or Chat removes its archive-based deadline but cannot recover deleted working files. Active or queued Runs prevent cleanup. These rules apply to existing working files. Chats already archived when the archive clock is introduced use the migration time for that clock, without extending an earlier inactivity deadline. Already-expired files may be removed without another checkpoint. Deleting working files does not delete product records or transcripts. This is not a permanent-backup or crash-recovery guarantee. Workspace deletion removes checkpoints, with unreferenced bytes removed by background cleanup. Confirmed local caches become eligible for immediate removal for completed, cancelled or archived Tasks and archived Chats. Other confirmed idle caches are removed after seven days. Offline or customer-controlled copies may remain. A later Run discards an unconfirmed crash copy and restores the last committed checkpoint.
Downstream model providers through OpenRouter
Task Machine's managed model catalog can route a request through OpenRouter to the model developer selected by the customer or by an approved routing policy. Examples of model developers available through OpenRouter include Anthropic, OpenAI, Google, xAI, Moonshot AI, MiniMax, Alibaba through Qwen, DeepSeek, Mistral AI, Zhipu AI through Z.ai, Meta, and Particle AI. The downstream provider depends on the selected model and OpenRouter's applicable provider arrangements.
The selected provider receives the prompt, context, tool definitions and calls, generated content, and request metadata needed to answer the request. Available developers and models can change. Task Machine updates this schedule when the approved provider set changes. Customer-owned coding-tool and model-provider accounts used by Local Workers are outside the managed AI controls. Task Machine-managed workspace AI features remain subject to the workspace requirements.
Task Machine also uses models from the following providers through OpenRouter for its own features. Each request uses a pinned model, requires zero-data-retention routing and does not fall back to another model. The cases and content listed below apply to these providers as a group, and this schedule does not assign a case or content to a particular provider or model.
TypeSafe AI, Inc.
Perplexity AI, Inc.
Liquid AI, Inc.
- Liquid AI states that it hosts and operates its services in the United States.
Microsoft
- Transcribes optional voice input.
Mistral AI
- Transcribes optional voice input.
Cases for which these models may be used
These models may be used for the following cases, and any of the providers above may serve any case. This schedule does not assign a case to a particular provider or model.
- Agent request checks, before an Agent asks a person a question or for an approval.
- Tool-call checks, before an Agent's tool call runs.
- Memory checks on proposed and saved workspace memory notes.
- Instruction checks on Playbook briefs, requirements and Run instructions.
- Writing checks on messages an Agent writes before they are published.
- Plan checks on proposed plans and Task acceptance requirements.
- Outcome and code review checks on submitted work and changed repository files.
- Other structured checks and suggestions that return typed answers.
- Optional voice dictation.
Content that may be sent
Depending on the case, requests can send bounded excerpts of Task and plan text, comments, Inbox context, memory evidence, workflow inputs, review feedback, Playbooks and their briefs, Tasks and their acceptance requirements, submitted work with its files and Git changes, Library documents, pull-request status, code review rules, Agent messages and requests with their answer choices, tool inputs and recent tool results, the names of the tools, Connectors and credentials available to an Agent, Run instructions and guidance, earlier decisions, recent Human messages from other Tasks, saved workspace memory notes, files, browser pages, Run transcripts, working directories, resolved file paths and repository details.
Every request carries typed questions and request and usage metadata, and returns typed answers.
Voice dictation sends short audio phrases, vocabulary hints containing Task Machine terms and names of workspace resources the person can access, and request metadata. Task Machine does not store voice recordings or a separate transcript history. Dictated text becomes ordinary workspace content only if the person submits it.
Customer-selected services
A workspace may connect third-party MCP servers, communication systems, source hosts, coding tools, model providers, public-web services such as Exa, and other services. Those providers process data because Customer selected and instructed the connection. They are not Task Machine subprocessors when Customer contracts with and controls them directly. Their own terms, regions, retention, and security practices apply.
Independent controllers and recipients
Stripe and Link: the Link entity identified at checkout is merchant of record for Stripe Managed Payments. Link processes payment methods, transaction, billing-country, tax, fraud, refund, dispute, and support data as an independent controller under the checkout and Link terms. Link is not a subprocessor for workspace content.
Cloudflare: separately acts as controller when using Turnstile and Challenge Platform signals to improve bot detection under its Turnstile Privacy Addendum. This is distinct from its processor role in providing our security and delivery services.
Google: when a person chooses Google authentication, Google receives the request and account information and acts under its own terms and privacy notice.
Task Machine may also disclose limited information to professional advisers, public authorities, courts, or transaction counterparties where lawfully necessary.
Changes and objections
Task Machine updates this schedule and gives active DPA customers at least 15 days' advance notice through their recorded contact before a material new subprocessor starts processing Customer Personal Data, unless urgent security, legal, or service-continuity circumstances require a shorter period. A customer may object during that period on reasonable data-protection grounds as described in the Data Processing Addendum.
Questions, objections, or requests for current provider documentation can be sent to [email protected] . Read the Data Processing Addendum.