Biclever Agent: security and IT review
What the agent can and cannot access, where credentials live, and what is logged. Written for the IT and security people reviewing a pilot.
Architecture
Each customer gets a private workspace. A mail desk with fixed rules and no AI talks to your team, and the agent works sealed off from the outside. The Biclever Gateway, installed in your own network, holds the passwords to your BusinessObjects system and reaches the agent through an encrypted relay.
Credentials
- The AI never sees your passwords. With the Biclever Gateway, the credentials stay in your network. With a direct connection, they are held by a separate gatekeeper component that logs in on the agent’s behalf.
- Each task gets its own fresh session.
How the agent connects to your system
There are two ways, depending on how your BusinessObjects system is reachable.
- Biclever Gateway (recommended). A small service you install on a server in your own network. It holds the BusinessObjects credentials, works with Web Intelligence through the BusinessObjects REST API and with universes through the Semantic Layer SDK, and writes every request to an audit log. The gateway sets up an encrypted channel to the agent, so the credentials and the connection stay on your side.
- Direct connection. If your BusinessObjects system is reachable from the internet, the agent can connect to it directly with the service account.
The Biclever Gateway in detail
- What it is. A Java service for a Windows or Linux server. On Windows it installs from an MSI and runs as a Windows service under its own low-rights account.
- No inbound ports. The gateway only makes outbound HTTPS calls to one fixed hostname, which we give you. It works through corporate proxies, including firewalls that inspect HTTPS.
- End-to-end encrypted. The gateway and the agent talk through a relay on Google Cloud in Belgium (europe-west1). The relay passes encrypted messages and can’t read requests or answers; it sees only routing details such as sizes and timing.
- You decide what it may do. The rules for each BusinessObjects system are in a local file on your server and can’t be changed remotely. A system without rules is read-only.
- Audit log on your server. One line per request: what was called, on which system, and the result. It never contains report content, passwords or logon tokens. Kept 90 days by default.
Supported versions: BI 4.3 SP3 and later, including BI 2025.
Access and the service account
The agent uses a dedicated service account with the rights you choose. Its access is restricted to one folder, so it can’t change reports outside it.
| Environment | In a pilot |
|---|---|
| Test | Yes, restricted to one folder |
| Production | Yes, if you prefer, restricted to one folder |
Finance teams often want the agent on production, where their reports are. The folder restriction and working on copies apply either way.
Who can use the agent
You decide who may write to the agent. Mail from anyone else, or from a sender address that can’t be verified, is ignored. Adding or removing a user later takes one email to us.
Protecting original reports
By default, the agent works on a copy. If you explicitly request a change to the original, it makes a backup first.
Data handling and hosting
Each customer has a dedicated agent with its own mailbox, memory and system access, running in an isolated workspace.
- Where it runs. The agent runs on a dedicated server in Norway.
- AI processing. Requests are processed by Anthropic’s Claude API, under Anthropic’s commercial terms. Anthropic doesn’t use API data to train its models.
- Email. The agent’s mailbox is on Google Workspace, which doesn’t use customer data to train models.
- Retention. Email threads and work files are deleted 90 days after the last message in a thread. The agent’s memory (its notes about your system and its task journal) is kept while you’re a customer and deleted within 30 days after the contract ends. You can ask us to delete everything at any time.
- Self-hosted. On request, the agent can run as a container in your own environment.
Who handles data for us
| Provider | What for |
|---|---|
| Anthropic | AI processing (Claude API) |
| Email (Google Workspace) | |
| Gateway relay (Google Cloud, Belgium), encrypted traffic only | |
| Stripe | Payments and invoices |
| GitHub | This website (GitHub Pages) |
| Web3Forms | The website’s pilot and contact forms |
A data processing agreement (DPA) is available on request.
Logging and traceability
Every system call is logged and every task is recorded in the agent’s journal, so past work can be found, reviewed and repeated.
Limits
AI can make mistakes. Important changes and results may require human review before they can be relied upon in production.
The agent handles everyday BusinessObjects work well, but some reports are very complex, and we can’t guarantee it handles them efficiently. When it can’t finish a task, it explains what’s missing and keeps a note. With your permission, we review those notes to keep improving it.
Questions from your IT team? Write to pilot@biclever.com, or apply for a pilot.