Skip to content

Twin.so AI Agents

Leveraging Twin.so AI Agents with Tallyfy

Twin.so offers sophisticated AI agents designed to automate complex operations by directly interacting with web applications through a browser, mimicking human behavior. Developed by Twin Labs (founded 2024), Twin.so has achieved significant real-world deployment success, including their “Invoice Operator” serving 500,000 European SMBs through their partnership with Qonto and OpenAI. This capability makes Twin.so a strong candidate for integration with Tallyfy, especially for automating tasks where traditional API-based automation is unavailable or impractical.

Important guidance for AI agent tasks

Your step-by-step instructions for the AI agent to perform work go into the Tallyfy task description. Start with short, bite-size and easy tasks that are just mundane and tedious. Do not try and ask an AI agent to do huge, complex decision-driven jobs that are goal-driven - they are prone to indeterministic behavior, hallucination, and it can get very expensive quickly.

Understanding Twin.so AI Agents

Twin.so’s AI agents, often referred to as “skilled agents,” are built to understand goals and autonomously navigate and interact with web interfaces. They are not reliant on pre-existing APIs for the applications they automate.

Key characteristics of Twin.so agents include:

  • Browser-Based Interaction: Agents operate directly within a web browser (typically a remote Chromium-based browser session), controlling the interface to perform tasks.
  • Natural Language Goals: You provide the agent with a starting URL and a goal defined in natural language. The agent then interprets this goal to carry out the necessary actions.
  • Production-Scale Deployment: Successfully deployed at enterprise scale, with Invoice Operator handling millions of invoices monthly across 500,000 European SMBs through Qonto.
  • OpenAI CUA Integration: Twin was selected as one of 15 companies to alpha-test OpenAI’s Computer-Using Agent (CUA) model, the same technology that powers OpenAI’s Operator.
  • Adaptive and Robust: Twin.so emphasizes that its agents can adapt to changes in website layouts and are more robust than brittle RPA scripts. They reason in real-time at each step.
  • Secure Access: Agents can use your private access key or credentials stored securely in Twin.so Vaults to automatically log into authenticated apps on your behalf.
  • Enterprise Infrastructure: They provide scalable, low-latency browser infrastructure designed for enterprise-grade security and large-scale operations.
  • Performance Leadership: Latest generation Twin A3 achieves 6-second latency per step, $0.03 cost per step, and 84% accuracy - representing industry-leading efficiency in browser automation.

How Twin.so Works in Detail

Twin.so’s automation capability is centered around its Twin Agent and proprietary Twin Model (also known as Twin’s Action Model).

  1. Task Initialization: A task for the Twin Agent is defined by two primary components:
    • A starting URL: The webpage where the agent begins its work.
    • A goal: A natural language instruction describing what the agent needs to accomplish.
  2. Execution Environment: Tasks are executed in secure, remote browser sessions controlled by Twin.so with specialized Kubernetes infrastructure maintaining latency below 50-100ms even at high volumes.
  3. The Twin Model - Multimodal Understanding and Action: This is the core AI powering the agent. It’s enhanced by OpenAI’s CUA model and optimized for customer use cases:
    • Perceive Web Environments: Understand the layout, elements (buttons, forms, text), and context of a wide variety of websites and web applications.
    • Decide Optimal Actions: Based on the current state of the web environment and the overall goal, the model determines the most appropriate next action.
    • Translate to Browser Instructions: The model’s decision is translated into machine-readable instructions that are then executed in the remote browser.
  4. Performance Evolution: Twin has dramatically improved over three generations:
    • Twin A1: 35s latency/step, $1.10 cost/step, 54% accuracy
    • Twin A2: 19s latency/step, $0.15 cost/step, 68% accuracy
    • Twin A3: 6s latency/step, $0.03 cost/step, 84% accuracy
  5. Collaboration with Leading Models: Twin leverages OpenAI’s CUA model and collaborates with OpenAI’s research team to improve capabilities for specific customer use cases.
  6. Output Generation: Upon completing the task, the agent returns an output with structured results, summaries, or confirmations of task completion.

Real-World Success: Invoice Operator

Twin.so’s first major deployment demonstrates the platform’s enterprise readiness:

  • Partnership: Built in collaboration with Qonto (European fintech) and OpenAI
  • Scale: Serves 500,000 SMBs across Europe
  • Capability: Automatically retrieves invoices from vendor portals, even those requiring login credentials
  • Performance: Successfully retrieved 500 invoices from 50 different service providers in under 10 minutes during testing
  • Impact: Reduces manual invoice collection time from hours to minutes per business

Getting Started with Twin.so for Tallyfy Integration

  1. Access Twin.so Platform:

    • Contact Twin.so for enterprise access or explore their platform capabilities through their website. Twin.so focuses on enterprise partnerships rather than self-service onboarding.
  2. Understand Enterprise Integration:

    • Twin.so typically works directly with enterprise customers to build specialized agents tailored to specific industry needs and workflows.
  3. Obtain API Access:

    • For direct API integration with Tallyfy, work with Twin.so’s team to establish API access and understand their integration patterns.
  4. Secure Credentials Setup:

    • Use Twin.so’s Vaults feature to securely store credentials for websites that your automated tasks will need to access. The agent can then use these vaulted credentials via your access key.
  5. Identify Tallyfy Tasks for Automation:

    • Review your Tallyfy processes and identify specific tasks that involve repetitive web browser interactions suitable for Twin.so’s capabilities.
  6. Define Integration Architecture:

    • Work with Twin.so’s team to architect the integration, which may involve:
      • Twin API: Direct API calls from Tallyfy webhooks or middleware
      • Custom Webhooks: Twin.so calling back to Tallyfy upon task completion
      • Enterprise Connectors: Purpose-built integrations for specific use cases

How Tallyfy Integrates with Twin.so (Example Scenario)

Tallyfy Task: “Fetch Monthly Invoice from OpenAI for Expense Report”

  • Inputs from Tallyfy Form Fields:
    • Service Provider: OpenAI
    • Invoice Month: February 2025
    • Expected Amount: $92.10
    • Login Credentials Vault: (Reference to Twin.so secure vault)
  • Integration Steps (Enterprise API):
    1. When this Tallyfy task becomes active, a Tallyfy webhook triggers an API call to Twin.so.
    2. The API call to Twin.so would be to Start Task with parameters like:
      • starting_url: https://platform.openai.com/account/billing
      • goal: “Log into OpenAI using vaulted credentials. Navigate to billing section. Find and download the invoice for February 2025 with amount $92.10. If not found, report as ‘Invoice not available’.”
      • access_key: Reference to securely stored OpenAI credentials.
    3. Twin.so agent executes the task in their enterprise browser infrastructure.
    4. The Tallyfy integration monitors task completion through Twin.so’s API or webhook callbacks.
    5. Once the Twin.so task is complete, the API response includes:
      • Task status (e.g., “completed”, “failed”).
      • Retrieved invoice file or download link.
      • Structured data about the invoice (amount, date, invoice number).
    6. The integration updates the Tallyfy task with the invoice attachment and extracted data, allowing the process to continue to expense report generation.

Use Cases for Tallyfy and Twin.so

  • Automated Invoice Collection: Retrieve invoices from hundreds of vendor portals automatically, as demonstrated with Qonto partnership.
  • Supplier Portal Management: Automatically check order status, delivery updates, and compliance documents across multiple supplier websites.
  • Customer Data Synchronization: Extract customer information from various CRM systems and update consolidated databases.
  • E-commerce Operations: Monitor product listings, competitor pricing, and inventory levels across multiple marketplaces.
  • Compliance Reporting: Automatically gather required documentation from regulatory portals and compliance systems.
  • Financial Data Aggregation: Collect account balances, transaction summaries, and financial statements from multiple banking and financial service providers.

Benefits

  • Enterprise-Proven Reliability: Demonstrated success at scale with 500,000+ users across major European markets.
  • Advanced Technology Integration: Benefits from OpenAI’s latest CUA technology and dedicated optimization.
  • Exceptional Performance: 6-second latency and 84% accuracy in latest generation represents industry-leading capabilities.
  • No API Dependencies: Automate tasks on any web application, regardless of API availability.
  • Enterprise Security: Purpose-built infrastructure with SOC2 Type 2 compliance and GDPR adherence.
  • Leverage Tallyfy’s Strengths: Use Tallyfy for overall process definition, human task management, approvals, and tracking, while delegating specific browser automation to Twin.so.

Potential Considerations

  • Enterprise Focus: Twin.so typically works with enterprise customers rather than offering self-service options, which may require higher minimum commitments.
  • Custom Integration: Integration with Tallyfy would likely require custom development work and coordination with Twin.so’s technical team.
  • Specialized Use Cases: While highly capable, Twin.so appears to focus on specific domains (like invoice processing) rather than general-purpose automation.
  • Cost Structure: Enterprise-focused pricing may be significant, though the high accuracy and low latency may justify costs for high-value processes.
  • Geographic Focus: Current major deployments are primarily in European markets, though this may expand.

By combining Tallyfy’s process management capabilities with Twin.so’s enterprise-grade AI agents, organizations can achieve sophisticated automation for complex, multi-step business processes that were previously impossible to automate effectively.

Computer Ai Agents > AI Agent Vendors

The Computer AI Agent market has rapidly matured in 2025 with enterprise-ready leaders like OpenAI Operator Claude Computer Use and Twin.so alongside open-source innovations such as Skyvern and Manus AI offering various approaches to autonomous web-based task automation that can integrate with Tallyfy workflows.

Integrations > Computer AI Agents

Computer AI Agents work with Tallyfy by providing intelligent automation capabilities that can perceive digital environments and execute complex tasks while Tallyfy serves as the orchestration framework that provides step-by-step instructions defines inputs and outputs establishes guardrails and ensures transparent trackable execution of AI-driven business processes.

Vendors > OpenAI Operator

OpenAI Operator is an AI agent launched in January 2025 that performs web-based tasks by interacting with browser interfaces like a human and can be integrated with Tallyfy processes to automate mundane web interactions such as form filling online ordering and booking reservations through natural language instructions.

Vendors > Skyvern AI Agents

Skyvern is an open-source browser automation platform that uses LLMs and computer vision to achieve 85.8% performance on the WebVoyager benchmark through its advanced Planner-Actor-Validator architecture and can integrate with Tallyfy to automate web-based tasks within business processes using natural language prompts.