Skip to contentNew Learn why APIs are the future Read the article →

Developer & security FAQ

Technical answers for production portal APIs

How Zatanna manages changing credentials, session-sensitive request flows, anti-bot controls, production traffic, and the latency and reliability tradeoffs of replacing headless browsers with request-layer APIs.

YYC W26 · Built for production request workflows
Classical engraving of a scholar studying at a desk, illustrating Zatanna's editorial law-meets-AI aesthetic

Architecture at a glance

Request layer, not the rendered UI.

Zatanna reconstructs the HTTP request flow a portal already uses, then exposes the workflow behind one stable endpoint.

01

State

Managed authentication, MFA, cookies, session refresh, and ordered requests.

02

Resilience

TLS-aware request behavior, session continuity, change detection, and repair.

03

Output

Structured responses without rendering and parsing a full page for every call.

FAQ

Common questions

Zatanna separates the API contract your agent calls from the portal session underneath it. The integration manages stateful authentication, cookies, MFA steps, and session refresh as part of the request flow. When a portal rotates a token or expires a session artifact, Zatanna refreshes or re-establishes that state without requiring the agent to orchestrate credentials. Exact handling depends on the portal; vaulting, rotation cadence, and customer-specific controls are confirmed during security review.

Frontend-only changes usually do not matter because Zatanna is not selecting and clicking visual elements. If the backend contract, login sequence, or session behavior changes, Zatanna's observability layer detects the failing request path and the integration is repaired. This narrows the maintenance surface; it does not mean an upstream system can never introduce a breaking change.

Modern anti-bot systems can combine TLS handshake characteristics, headers, cookie continuity, request timing and order, IP reputation, and browser signals. Zatanna works at the request layer with TLS-hardened, session-aware request sequencing, reproducing the valid network behavior of the workflow rather than depending on a default headless-browser fingerprint plus DOM stealth patches. No integration layer can promise a universal bypass: feasibility and controls are evaluated per portal.

Not as a blanket promise. Some flows can preserve a valid authenticated session or incorporate an approved human step; other controls may make a workflow unsuitable. Zatanna evaluates the portal, the customer's authorization, and the required action before defining the production path.

Capture should be scoped to the authorized workflow and environment needed to reconstruct request behavior, not treated as unrestricted observation of a user's activity. Zatanna's published privacy policy describes data minimization, access controls, retention practices, security safeguards, contractual restrictions, and human oversight. It also states that personal information is not sold and is deleted or anonymized when no longer needed. Exact retention, residency, access, and customer-specific handling should be documented during workflow and security review.

The objective is to preserve the minimum state and request knowledge needed to operate and repair the integration, not to retain an indiscriminate traffic archive. The data required varies by workflow and portal. Teams with sensitive fields or regulated data should define payload scope, logging, retention, and access requirements during review before production traffic is enabled.

Playwright and Puppeteer control a browser: create a page or context, navigate, render, execute JavaScript, wait for page state, and interact with the DOM. Zatanna calls the reconstructed request flow directly and returns structured data. Zatanna's existing product benchmark is roughly 200–500 ms for a supported workflow, while browser automation can add seconds before the upstream action finishes. This is indicative, not a universal SLA: network distance, portal response time, authentication, rate limits, and workflow length still determine end-to-end latency.

It removes UI-specific failure points: selectors, layout changes, animation timing, page hydration, rendering differences, and browser binary drift. A moved button does not change the underlying request contract. Request-layer integrations can still break when an upstream API, authentication sequence, or anti-bot policy changes; Zatanna addresses that smaller failure surface with monitoring, change detection, and repair.

Yes. Use a headless browser for end-to-end testing of an application you control, visual validation, workflows whose output genuinely depends on rendered UI state, and low-volume exploratory automation. Use Zatanna when a repeated third-party portal workflow needs to behave like production infrastructure behind a stable endpoint.

Measure the same real workflow across both approaches: cold and warm end-to-end latency, success rate over repeated runs, session-expiry recovery, behavior after a harmless UI change, concurrency, upstream rate-limit handling, and engineering time spent on selectors, waits, proxies, and browser/runtime upkeep. Report the portal and test conditions with the result; one synthetic number without context is not a useful reliability claim.

Bring one blocked workflow. We'll turn it into a production API.

30-minute working session