Figranium v0.15.0 introduces built-in CAPTCHA solving for Agent mode, covering reCAPTCHA v2, reCAPTCHA v3, hCaptcha, and Cloudflare Turnstile.

That means a browser task can now encounter a supported challenge, solve it as part of the same execution, inject the resulting token, and continue the workflow without handing control off to a separate manual step.

Yes, this also means the robot can now click “I’m not a robot.”

Two ways to solve CAPTCHAs#

You can handle challenges explicitly with the new Solve Captcha block, or enable Auto-Solve Captchas at the task level.

The explicit block is useful when you know exactly where a CAPTCHA appears in a flow. You can optionally specify the CAPTCHA type and selector, or let Figranium auto-detect the challenge from the page.

Auto-Solve Captchas is designed for less predictable workflows. When enabled, Figranium checks for a supported challenge after navigate, click, and type actions. If no challenge is present, it silently does nothing.

Auto-solve is off by default, so tasks that do not need CAPTCHA handling do not pay the extra latency.

Local solving is built in#

The current solver is resource-adaptive and can run locally using the task's own browser session.

For image-based reCAPTCHA v2 and hCaptcha challenges, Figranium selects a model tier based on available memory:

• OWL-ViT on hosts with 2–7.99 GiB of effective memory
• Florence-2 on hosts with 8 GiB or more

Turnstile and checkbox-only test-key flows can be handled without loading an image model.

Model weights are fetched on first use into persistent storage rather than being baked into the Docker image, and only the active tier is retained at steady state.

Local image solving requires at least 2 GiB of effective cgroup memory.

Optional remote fallback#

Figranium can also use a YesCaptcha/AntiCaptcha-compatible remote endpoint. If configured, the remote route is tried first and the local solver can take over on failure.

This makes it possible to choose between fully local solving, a remote provider, or a combination of both without changing the task itself.

Sensitive browser-context forwarding is disabled by default. Proxy and browser-context forwarding are separate opt-ins for users who explicitly need them.

What the solver returns#

When you store the result in a variable, Figranium exposes useful execution details including:

• whether solving succeeded
• the detected challenge type
• total duration
• the provider that produced the token
• the local model and device, when applicable
• per-route attempt results

That makes CAPTCHA handling observable enough to log, branch on, and debug like any other part of a workflow.

More than CAPTCHA support in v0.15.0#

CAPTCHA solving is the headline feature, but v0.15.0 also includes several other changes:

• a new Do Nothing block for explicit no-op steps
• a generated openapi.json covering the REST API surface
• a clearer task API trigger panel that shows the full origin-aware POST endpoint and method badge
• a fix for executeTaskById crashing when external callers omit Content-Type: application/json
• a redesigned All Captures screen using a simpler two-column grid

The release also removes an older unused CAPTCHA handoff module that was never wired into the runtime and is now superseded by the new solver.

Available now#

CAPTCHA solving is available in Agent mode in Figranium v0.15.0.

For workflows where challenges appear consistently, use the explicit Solve Captcha block. For workflows where a challenge may appear unpredictably, enable Auto-Solve Captchas in the task's Behavior settings.

As always, only automate sites you own or have permission to automate. CAPTCHA solving on third-party services can violate their terms or applicable rules.