How to Research Private Companies With Codex and Akta.pro

To research private companies with Codex, define the evidence you need, resolve each company to its official domain, retrieve only the minimum structured data required for the decision, and verify important claims against primary sources. Aident Loadout gives Codex access to Akta.pro for the structured search and enrichment steps without putting an API key in your repository.

The useful output is not a generic company summary. It is one reviewable record per intended company, with the exact entity, source links, dates, unknown fields, and a confidence-based review status.

This workflow is for screening a finite company list for sourcing, sales research, or market mapping. The private-company news guide owns ongoing news monitoring after the entities are known.

Define the Research Contract Before Calling an API

A request such as “research these companies” is too open-ended to review or price. Start with a table that connects every field to a decision.

Do not make a field mandatory merely because a provider can return it. A smaller contract is cheaper to run and easier to verify.

Step 1: Set Up Aident Loadout

Give Codex the canonical setup instruction:

Follow https://aident.ai/SETUP.md

Then confirm authentication and Vault readiness:

aident account auth status
aident vault vault --action status

Ask Codex to discover the current Akta.pro company-search and enrichment Actions and inspect their live schemas. Do not copy an internal Action identifier into a permanent prompt. Catalog versions, inputs, platform access, and pricing can change.

Akta.pro describes its service as structured, source-attributed intelligence covering more than 20 million private companies. Treat that as the provider's coverage statement, not proof that every company or field will be present.

Step 2: Resolve One Company Before Enrichment

Search one input at a time and stop before the priced enrichment step:

Find the current read-only Akta.pro company-search Action in Aident Loadout. Search for [company name or domain]. Return candidate names, official websites, stable identifiers, and distinguishing metadata. Do not enrich a candidate until its official domain matches the intended company.

On August 24, 2026, a bounded live search for Anthropic returned 25 candidates, including more than one plausible record. That single test does not measure the catalog's general accuracy, but it demonstrates why “take the first match” is not a safe resolution rule.

Prefer an exact official-domain match. If the input is a subsidiary, product, or former name, preserve that relationship rather than silently converting it into the parent company. When two candidates remain plausible, mark the record needs review and spend nothing on enrichment.

Expected result: one reviewed company identity or an explicit unresolved state.

Step 3: Request the Minimum Enrichment Sections

Once the identity is reviewed, ask for only the sections needed by the research contract:

Preflight the current Akta.pro company-enrichment Action for the reviewed company. Request only the firmographic, digital-presence, and industry sections. Show the exact quote and the section identifiers accepted by the live schema. Do not execute until I approve the quoted spend.

In a live preflight on August 24, those three sections produced a 14-credit quote. The priced call was not executed for this article. Re-run preflight before every workflow because available sections and prices can change.

Keep provider output separate from your conclusion:

{
  "inputCompany": "Example company",
  "officialDomain": "example.com",
  "providerFacts": [],
  "primarySourceChecks": [],
  "fitDecision": "needs review",
  "confidence": "low",
  "unknownFields": []
}

Do not infer ownership, revenue, funding, or niche fit from a field that was not returned.

Step 4: Add Signals Only When They Change the Decision

Enrichment is a base record. Optional signals should answer a named question, not make the report look fuller.

These are dated preflight examples, not permanent prices or proof that the data will be available for a specific company. Inspect the current schema, show the quote, and require approval before each priced call. This guide explains Loadout credits, BYOK, and Action pricing.

Use the dedicated private-company news workflow when news monitoring becomes a recurring job. Keep the screening canonical focused on entity resolution, fit, and evidence completeness.

Step 5: Corroborate Claims With Primary Sources

Source metadata makes a fact inspectable; it does not make every downstream conclusion correct.

For each high-impact field:

1. Open the cited source and confirm it refers to the resolved company.

2. Record the publisher and date.

3. Prefer the company's site, a regulator, or another primary record for legal identity and ownership.

4. Distinguish a direct fact from a provider classification or your fit judgment.

5. Mark stale, conflicting, and inaccessible evidence instead of averaging it into a confident answer.

Private-company records are often incomplete. unknown is a valid output when the evidence cannot support a decision.

Step 6: Batch Only After the Single-Company Run Passes Review

A public Reddit request in August 2026 described a desired workflow for screening 500 to 2,000 private companies per day for M&A sourcing, including independence, niche fit, confidence, and citations. That is useful evidence of the operational job, but it is not evidence that this workflow has been validated at that throughput.

Before scaling, test a representative batch and record:

• exact-domain resolution rate;

• unresolved and duplicate candidate rate;

• percentage of required fields with dated sources;

• reviewer acceptance rate;

• false-positive fit rate;

• credits per accepted company; and

• median review time per record.

Process a bounded batch with limited concurrency. Preserve the original row order and one result status per input. Retry only transient transport failures; do not retry ambiguous identities or repeat a priced call whose outcome is uncertain.

Step 7: Produce a Review Queue, Not an Automatic Decision

Ask Codex for a compact output:

Build a review table from the resolved and verified company records. Preserve the input company, official domain, canonical identifier, fit decision, supporting facts with source URLs and dates, unknown fields, confidence, quoted and actual credits, and review status. Put unresolved identities and conflicting evidence in a separate exception queue. Do not contact companies, update a CRM, make an investment recommendation, or execute another priced Action.

Success is measurable when every input has exactly one terminal status, every accepted claim is traceable to a dated source, ambiguous entities are excluded from enrichment, and the cost per accepted record is visible.

Common Failure Modes

Similar Names Resolve to the Wrong Company

Compare the official domain and distinguishing metadata before enrichment. A stable identifier is useful only after it is attached to the right organization.

The Report Fills Missing Fields With Guesses

Require an evidence URL and date for each important claim. If a field is absent, write unknown and route it for review.

The First Batch Costs More Than Expected

Start with the base sections, preflight optional signals separately, and record credits per accepted company. A large result limit does not improve a weak research contract.

News Monitoring Takes Over the Screening Job

Use recent news only when it changes the current decision. Move recurring event tracking to the separate news-monitoring workflow.

The Agent Treats Research as an Investment Decision

Keep the output descriptive and reviewable. A company data record is not financial, legal, or investment advice.

Reusable Codex Prompt

Paste this after setup and replace the bracketed values:

Use Aident Loadout to research [bounded company list] for [explicit inclusion rule]. Confirm account authentication and Vault readiness. Discover and inspect the current read-only Akta.pro company-search and enrichment Actions. Preserve every input, resolve each company to its exact official domain, and stop unresolved identities before any priced call. Preflight the minimum enrichment sections needed for [required fields], show the quote, and wait for approval. Keep provider facts, primary-source checks, and fit judgments separate. Preserve source URLs and dates, use unknown for missing evidence, and route ambiguous or conflicting records to an exception queue. Report credits per accepted company. Do not contact anyone, write to another integration, or make an investment decision.

Set up Aident Loadout and research one private company with a review gate.

Sources

• Akta.pro private-company data and signals API

• Aident Loadout setup

• Reddit request: automate screening 500 to 2,000 private companies per day

• How to Monitor Private Company News With Codex

• How Aident Loadout Credits, BYOK, and Action Pricing Work

Refresh this guide when the Akta.pro search or enrichment schema, provider documentation, Loadout setup flow, pricing behavior, or linked canonical workflows change.