Homula Launches AI Accounting Agent on agens to Automate Invoice Downloads from Multiple Websites Without Exposing Passwords to External Systems or LLMs

Homula Inc. (Headquarters: Akasaka, Minato-ku, Tokyo) announces the launch of a new solution on its AI agent execution platform 'agens (Agents)', designed to automate the retrieval of invoices from web portals. The solution is targeted at finance departments of large enterprises and invoice receipt outsourcing (BPO) service providers. Alongside the solution, Homula is launching an implementation support program that guides clients from requirements gathering to full production deployment.

While electronic invoice delivery is becoming widespread, the delivery channels remain unstandardized. Finance and BPO teams continue to manually log in each month to numerous different websites—unique to each vendor—to download invoices one by one. agens automates this process entirely within the client's own internal environment.

Unlike dedicated agents embedded in specific SaaS platforms, agens stands out with three key features: (1) It is not limited to invoice retrieval and can be freely customized; (2) It allows the use of LLMs already approved within the organization; and (3) It processes login credentials and invoice data without exposing them to external SaaS providers or the LLMs themselves. Furthermore, agens simultaneously extracts invoice contents into structured data and synchronizes it with downstream systems such as accounting software and ERPs, enabling end-to-end automation beyond just 'download'.

Background: Invoices are 'digital', but receipt remains manual

While the shift to electronic invoicing and e-invoicing standards like Peppol is accelerating, delivery methods remain fragmented. Each vendor uses different invoice portals and issuance systems, and a single recipient company may need to access dozens of different sites. As a result, finance staff must repeatedly log in each month using different URLs, IDs, and passwords, wait for invoices to be issued, and download them individually. Missed issuances due to timing mismatches, overlooked downloads, and subsequent inquiries to vendors are common. For BPO providers handling invoice receipt on behalf of clients, this 'manual download from websites' is one of the most labor-intensive processes.

Two Limitations of Existing Approaches

Traditional solutions face two major limitations:

Continued manual work — Automation is difficult due to the diversity of sites, so monthly retrieval remains a manual, labor-intensive task.

Dependency on embedded agents in specific SaaS — This approach locks users into fixed use cases, models, and data flows. Recently, some invoice receipt SaaS platforms have introduced built-in automation agents. However, these solutions are limited to the vendor's intended use case, restrict users to the vendor's designated AI model, and require data to be processed on the vendor's cloud infrastructure. For large enterprises with strict data sovereignty requirements or BPO providers managing multiple clients' data, this creates contractual and security barriers.

agens' Solution: General-purpose harness × Model-free × Self-hosted

agens is not a fixed-purpose product but a general-purpose execution and control environment (agent harness) designed to safely delegate real-world tasks to AI. Invoice retrieval from websites is just one use case running on this platform.

Compared to embedded SaaS agents, agens differs in five key aspects:

Scope of application — Traditional embedded agents are limited to the vendor's intended invoice processing workflow. agens is not limited to invoice retrieval and can be applied to any PC-based task, with full customization.

Available LLMs — Traditional solutions lock users into vendor-specified models. agens allows organizations to freely select from their approved LLM APIs (e.g., Azure, AWS, GCP).

Data location — In traditional models, data flows to the vendor's cloud. agens operates entirely within the client's internal environment, and with proper configuration, data never leaves the organization.

Handling of login credentials — Traditional systems store credentials on external SaaS platforms. agens keeps credentials within the internal environment and executes logins without exposing passwords to the LLM.

Use in BPO (outsourced receipt) — Traditional SaaS requires clients to entrust customer data to external platforms, creating contractual constraints. agens enables secure, multi-client invoice retrieval within the BPO provider's own environment.

How it works — agens provides generative AI with an isolated sandbox (virtual PC), where the AI autonomously observes screens and constructs the sequence of actions—from login to download—and executes them. Once a retrieval sequence succeeds, it is saved as a 'skill' and can be rapidly and reliably reused in subsequent runs. All operations occur within the client's internal network, mimicking the actions of a human operator.

Beyond 'retrieval' — Connecting to downstream invoice processing — agens not only retrieves invoices but also extracts their contents into structured data in real time. Key fields such as issuer, amount, date, and line items are synchronized directly into accounting systems, ERPs, or existing invoice processing platforms through seamless data integration. This eliminates traditionally manual steps involving OCR or human data entry. agens supports MCP (Model Context Protocol), enabling integration with existing systems without vendor lock-in (compatibility and extraction accuracy may vary by site and document format).

Design that keeps login credentials away from 'external systems' and 'AI' — Site-specific IDs and passwords are securely managed within the client's internal environment. The agent's execution environment inputs credentials directly during login. Passwords are never included in the LLM's prompts or context, nor are they transmitted to the LLM provider. Combined with self-hosting, this ensures that neither authentication data nor invoice content is exposed to external SaaS providers or LLMs. Unlike cloud services that require centralizing multiple site credentials on an external platform, agens structurally minimizes the risk of data aggregation.

Two Deployment Models

① For enterprise finance departments — Automate web-based invoice retrieval entirely within the company's internal environment. Organizations can use their already-approved LLMs, making adoption feasible even in companies with strict AI usage policies.

② For invoice receipt BPO providers — Securely retrieve invoices on behalf of multiple clients within the provider's own internal environment. Once a retrieval 'skill' is built, it can be reused across similar sites, reducing incremental costs as the number of clients and sites grows. This capability can serve as a key differentiator in BPO service offerings.

Inquiries and Enrollment

Implementation Program: Full-cycle support from PoC to production

To avoid solutions that stall at the PoC stage, homula offers hands-on implementation support to guide clients from requirements analysis to full production deployment, using its LLM-native FDE (Forward Deployed Engineer)