PartLine

One request. Up to five supplier calls. One ranked decision surface.

PartLine sends one adaptive CALL-E task across up to five approved suppliers. Every conversation returns the same structured evidence: exact match, stock, quantity, price, ship date, cutoff, lead time and spoken support. Deterministic code then checks exact match claims against the original request and ranks the suppliers side by side.

This is a comparative decision workflow, not a collection of isolated call summaries. The result is one review surface that answers which supplier can ship the exact part in time and what evidence supports that conclusion.

The differentiator

PartLine keeps quoted supplier evidence separate from derived ranking. An exact match must agree with the original manufacturer and part number. An alternate carries its caveats. A supplier who deflects to email remains UNKNOWN instead of being flattened into a false no stock result.

That unresolved supplier is intentional. Ambiguity is a first class state that tells the buyer where human follow up is still required.

The evidence console

Run partline web to open the local evidence console. It puts the approved request, ranked supplier claims, spoken evidence, engineering caveats and buyer decision states on one screen.

The browser receives no API key and no full phone number. It cannot place calls. The console is a visible trust boundary as well as the product interface.

How it works

Validate one exact part request, quantity, deadline and approved supplier list. Render a no call preview with masked numbers and a request bound approval token. Send one multi recipient CALL-E task across up to five suppliers. Collect the same structured availability and shipping evidence from every conversation. Validate exact match claims, rank the candidates and route every alternate or unknown result to human review.

Why this matters

When a production line stops, inventory pages become guesses. Maintenance and procurement teams need to verify the exact manufacturer and part number, physical quantity, shipping cutoff and lead time. Repeating that fact finding call across suppliers consumes the hours when downtime is most expensive.

PartLine compresses those calls into one comparable evidence workflow without turning the agent into an autonomous buyer.

Technical implementation

PartLine is a Python 3.11 application with no third party runtime dependencies. It integrates the CALL-E Developer API at runtime and includes:

One to five recipient calling with strict overall and per recipient JSON schemas Deterministic exact match validation and evidence based ranking Dry run preview, purpose bound authorization and request bound approval Stable idempotency keys that protect against duplicate call batches Weekday calling window enforcement A responsive localhost evidence console Twelve safety focused unit tests and deterministic review fixtures

Live validation

On August 27, 2026, one authorized 103 second English CALL-E call ran against an owned test recipient. The AI disclosed that it was calling on behalf of PartLine and captured 20 units available, a price of $50 per unit, approximately five day fulfillment and no shipping cutoff.

The call completed with high confidence. No order, reservation, negotiation or commercial commitment occurred. The phone number and transcript are excluded from public materials.

Safety boundary

PartLine may gather timely facts. It may not order parts, reserve stock, negotiate, accept terms or approve an engineering substitution. Supplier statements remain evidence for a human purchasing decision.

Reusable Agent Skill

The merged application is available in PR 246.

The merged follow up contribution packages the workflow as a reusable Agent Skill with tested examples, safety guidance and fictional phone fixtures. Review it in PR 263.

What Codex contributed

Codex helped research the contribution catalog, reject overlapping concepts and focus PartLine on comparative supplier evidence. It supported the scope, product design, Python implementation, evidence console, safety tests, repository validation, public pull requests and authorized live CALL-E validation.

What comes next

The next version adds durable webhooks, ERP request import, supplier calendars and export of selected evidence into an existing purchase requisition. PartLine will remain evidence infrastructure, not an autonomous buyer.

Built With

• call-e
• css
• html
• javascript
• python