Company intelligence, inside your agent

Clearbit research that shows its working

Turn an ICP brief into a reviewable company shortlist, named buying roles, and source-aware context. Every output is designed for a human to inspect before outreach begins.

Inspect the workflow
INSTALLnpx -y @okki-global/okki-go-taroball
Sample task
Find cybersecurity firms hiring their first RevOps lead; return domain, role rationale, and evidence.
AI agent producing an auditable company research table

Operating specification

What the workflow does—and does not assume

Research inputs

ICP definitionIndustry, geography, employee band, technology clues, and exclusions written in plain language.
Buying rolesFunctions and seniority to investigate; no guessed personal email is treated as verified.

Evidence controls

ProvenanceSource context stays beside the field it supports so reviewers can challenge the match.
Human reviewOutreach remains a separate decision after relevance, consent basis, and contact quality checks.

Delivery behavior

OutputStructured findings prepared for inspection, enrichment, or an approved downstream workflow.

Decision trail

A prospect moves forward only when the evidence does

01

Define the claim

Translate “good account” into observable firmographic, technographic, and operating signals.

02

Gather candidate evidence

Collect company and role context while keeping uncertainty visible instead of filling gaps with confidence.

03

Resolve identity

Compare domain, company name, location, and role clues before accepting a record as the same entity.

04

Approve the next action

A human checks relevance and regional outreach requirements before any message enters a sequence.

Evidence grid

Four research jobs, each with an explicit limit

Find companies

Input
An ICP expressed as observable criteria.
Agent action
Build and narrow a candidate set.
Output
Domains with match rationale and source context.

Limit: A signal indicates fit; it does not prove buying intent.

Verify contacts

Input
Target function and seniority.
Agent action
Resolve likely decision-makers and verification status.
Output
Reviewable role and contact evidence.

Limit: Unverified details remain labeled, never promoted to fact.

Prepare context

Input
Approved account and research question.
Agent action
Summarize triggers, technology, and role relevance.
Output
Concise briefing with citations.

Limit: Public signals can be stale and require a recency check.

Create outreach

Input
Approved evidence and message constraints.
Agent action
Draft a relevant, restrained first touch.
Output
Human-review copy with opt-out language.

Limit: The workflow does not auto-send or guarantee replies.

Installation runbook

Copy, run, configure, then inspect

  1. 01

    Copy

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

    Success means the exact command is in your clipboard—not that software has run.

  2. 02

    Run

    Paste it into a supported terminal and review the package prompt. A normal success signal is a completed setup message without a shell error.

  3. 03

    Configure

    Add only the credentials required for your chosen data source. Keep API keys in environment secrets, not prompts, transcripts, or committed files.

  4. 04

    First result

    Ask for a small test set. Verify domain identity, source recency, contact status, and exclusions before expanding the task.

Questions before you run

Installation and research FAQ

No. Copying only places text in your clipboard. Installation begins after you paste and run the command in a terminal and accept any runtime prompt.

No. Research and draft preparation are kept separate from sending. A person should approve targeting, claims, opt-out language, and applicable regional requirements.

Use your runtime’s secret store or environment configuration. Never paste credentials into a shared prompt, source file, or public issue.

No. Results vary by market, data availability, verification method, and recency. Run a representative sample and record your own acceptance and bounce thresholds.