Website visitor intelligence for your AI agent

Turn account activity into a prospecting brief your team can audit

Leadfeeder researches which companies may be visiting, enriches the account context, and prepares buying-signal evidence without hiding attribution uncertainty or taking the send decision away from your team.

Inspect a sample result
AI agent reviewing website visitor and CRM buying signals
Source context attached to consequential fieldsChecked time visible before actionUnknown state preserved when identity is uncertain

Evidence before outreach

Four linked checks for account-level prospecting

01

Identify likely companies

Input first-party website activity and an approved domain scope. The agent assembles candidate organizations and keeps anonymous or low-confidence sessions explicitly unresolved.

02

Enrich firmographic context

Add industry, size, geography, technology, and CRM ownership from configured sources. Conflicting records remain visible rather than being silently overwritten.

03

Read buying signals

Summarize visited topics, recency, repeat activity, and account fit. Activity is a prioritization clue—not proof that a named person intends to buy.

04

Prepare human review

Produce a research brief with provenance, checked times, confidence notes, suppression context, and a clear approval gate before any external action.

Auditable preview

See the task, result, and evidence together

SAMPLE

Prospect result

Account: Northstar Systems

Observed theme: CRM enrichment and attribution

Priority: Review — repeat activity, identity unresolved

Next step: Confirm fit and account owner

Connection boundaries

A runtime matrix, not a partnership logo wall

Input sources

Website activityConnection: configured analytics or event source
Permission: read approved events
Status: Config-dependent

CRM account recordsConnection: API or controlled export
Permission: selected account fields
Status: Config-dependent

Research agent

Firmographic synthesisCombines configured evidence
Status: Workflow available

Buying-signal briefLabels inference and uncertainty
Status: Human review required

Output destinations

Review queueConnection: agent workspace
Permission: operator-controlled
Status: Runtime-dependent

CRM updateConnection: scoped write action
Permission: explicit approval
Status: Optional

Four-state runbook

Copy, run, configure, then inspect a first result

  1. 1. Copy

    npx -y @okki-global/okki-go-taroball

    Success: the exact text is in your clipboard. This is not installation.

  2. 2. Run

    Prerequisite: review package identity and use an approved Node runtime. Success: the installer completes without an unreviewed permission request.

  3. 3. Configure

    Prerequisite: scoped source and destination credentials in a secret store. Success: a test connection reads only approved fields.

  4. 4. First result

    Prerequisite: a small authorized sample. Success: the brief exposes sources, checked times, uncertainty, and a pending review state.

Installation and data questions

Does copying the command install anything?

No. Copying only places the exact command on your clipboard. You decide whether and where to run it.

Does a visitor signal identify a person?

Not necessarily. Company resolution and person-level attribution are different claims, and ambiguous activity must remain labeled as unknown.

Can the workflow send outreach automatically?

The recommended operating model keeps targeting, suppression, message approval, and sending under human control.

Which integrations are supported?

Compatibility depends on the selected agent runtime and approved connectors. Verify current documentation before granting credentials.