Configure Plataforma de Recepción Inteligente

What it does

This page covers the configuration tasks for Plataforma de Recepción Inteligente that only the administrator performs: document types, cycles, confidence threshold, AI model, and reminder policy.

Before you start

• Role: instance administrator.
• Access to Settings → Technical and to the Recepción Inteligente menus.
• Estimated time: 20–30 min for the initial setup.

Define expected document types

Each cycle expects specific types (national ID, activity certificate, financial report). The catalog is reused across cycles.

1. Open Recepción Inteligente → Configuration → Document types. If your instance is federated with the canonical documentation, the basic sector types are already loaded; do not duplicate them.
2. To add one, click Create and fill in Code (short, lowercase, no spaces; used by the AI classification and should not change), Name, Area (identification, formulation, execution, or justification), and Instructions for the AI agent. The more specific, the better: “document with CIF/NIF, legal name and registered address; one page, AEAT PDF”.
3. Save. The type becomes available for flows and cycles.

Official documentation

Attachment management on Odoo records is documented in Odoo 19 · Productivity.

Create a flow and a collection cycle

The flow is the template (types, recipients, tone). The cycle is a concrete instance of the flow for a given period.

1. Open Recepción Inteligente → Configuration → Flows and click Create.
2. Fill in Name, Recipient source (one of the six canonical values: general members, internal users, custom, project receiving partners, project funders, project team), Project if the source is project-based, and Style profile.
3. Under Expected types add the types from the catalog that will be requested. Under Reminder policy edit the Escalation rules (dedicated section below). Save.
4. Go back to Recepción Inteligente → Cycles, click Create, associate the cycle with the flow, and define Start date and End date.
5. Click Open in the header. The system generates a request per recipient with their unique link and sends the initial email.

Adjust the confidence threshold

The threshold decides when the system automatically validates a document and when it places it on hold.

1. For the default value, open Settings → Technical → System parameters and search for coop_documentation.classification_threshold. Its default value is 0.7.
2. To override it for a specific flow, open the flow and fill in Confidence threshold. A value of 0.0 means “use the system default”.
3. If you activate the portal with verification, adjust coop_documentation.verify_threshold (default 0.7) for the cross-check between the type declared by the recipient and the type detected by the AI agent.

How to choose the threshold

A low threshold (0.5) reduces manual work but increases the risk of incorrect validations. A high threshold (0.9) guarantees precision but leaves most documents on hold. Recalibrate after one or two real cycles.

Connect the AI model

The platform uses the AI model configured for the instance. Open Configuration → AI model and verify it has Type: chat (embedding models are filtered out). To prioritize a model per document type, associate it in the AI model field of the type when available. [VERIFY: ticket multi-modelo per-tipo pendiente.]

Configure the escalated reminder policy

When a request is late, the daily cron job decides whether to send a reminder and at which level. The policy lives in the flow as a short JSON.

1. Open the flow and go to Reminder policy.
2. Escalation rules accepts a list of three entries, one per level (1 friendly, 2 assertive, 3 urgent). Each entry carries tier (1, 2, or 3) and days_after_deadline (days after the deadline; a negative value triggers pre-alerts before the due date).
3. Canonical example: [{"tier": 1, "days_after_deadline": -3}, {"tier": 2, "days_after_deadline": 2}, {"tier": 3, "days_after_deadline": 7}] produces a friendly pre-alert 3 days before the deadline, an assertive one 2 days after, and an urgent one a week later.
4. An empty list ([]) means silence: the flow does not send automatic reminders. The cap is strictly 3 levels; extras are ignored and logged.
5. Save. The cron picks up the policy on its next run.

How reminders are drafted

The text is composed from the flow’s style profile and the level. A tone field inside the JSON is deliberately ignored: the system uses the hard-wired level–tone mapping.

Schedule and monitor the daily cron

Automatic sending runs as a scheduled Odoo action under Settings → Technical → Automation → Scheduled actions → Recepción Inteligente: send pending reminders. By default it runs every day at 08:00 UTC: keep a single daily run and adjust the time if you work in a different timezone. To audit a run, search the server log for lines containing coop_documentation._cron_send_pending_followups.

Troubleshooting common issues

• The cron runs but sends nothing. The flow has empty rules ([] is deliberate silence) or the requests have no email address.
• Reminders land in spam. The cron sender does not match the address authorized on the mail server. In DonDominio, Postfix, or SES it must match the Responsible of the cycle.
• The AI agent classifies with very low confidence. The Instructions of the type are too brief. Expand them with examples of the expected format.

Next steps

• Working with requests from Plataforma de Recepción Inteligente so the operator can validate what you are about to configure.
• Plataforma de Recepción Inteligente: what it is and what it does if you need to understand how the module fits within Futuh.
• Glossary for canonical terms.