Autonomous background automation that extracts phone numbers, job titles, companies, and signatures from every incoming message. Synchronized directly to your Google Workspace directory with strict JSON schema verification.
Mit freundlichen Grüßen
Michael Meixner, CISSP
Allgemein beeideter Sachverständiger
Computerforensic & more GmbH
Tel: +43 664 144 83 63
Fritz Schmerold-Gasse 36, Vösendorf
{
"is_contact": true,
"display_name": "Michael Meixner",
"email": "mmeixner@forensic.wien",
"phone": "+43 664 144 83 63",
"company": "Computerforensic GmbH",
"job_title": "Gerichtl. Sachverständiger",
"notes": "CISSP Certified, Vösendorf"
}Recorded over continuous production scans across real executive mailboxes
Continuous execution on live inbox with zero throttling interruptions.
Blended rate factoring bot skips, cache hits, and fresh LLM inference.
Accurately identified business entities from free-form email footers.
Zero reasoning token overhead with native Gemma 4 KV-cache caching.
How many emails can you process per hour with your setup?
Eliminates hallucination and broken JSON. The vLLM inference engine uses grammar-guided decoding so the LLM output is guaranteed to strictly adhere to the Zod / Pydantic data schema.
In our live benchmark, 472 of 655 emails were resolved via instant cache lookups. Once a sender's contact information is verified, subsequent emails skip LLM calls entirely.
When deployed self-hosted, your emails and extracted contact records never leave your own servers. All inference runs locally on your private accelerator via WireGuard/Tailscale.
Run it self-hosted on your own hardware appliances, or connect with 1-click as a managed cloud service.
Deploy our declarative NixOS/Docker dendritic stack directly on your infrastructure with 100% on-premise inference.
Requires modern GPU accelerator with 16GB+ VRAM for Gemma 4 FP8 (e.g., NVIDIA GB10 Grace-Blackwell or RTX 4090 / A5000). Check certified hardware specs on gb10.work.
No hardware required. We manage the Gemma 4 cluster, OAuth pipelines, and hourly syncs for your entire team.
In typical production mailboxes, you can process 1,100 to 1,500 emails per hour in single-threaded mode. For massive archival backfills (100k+ emails), enabling parallel workers against our high-concurrency vLLM cluster scales throughput to over 12,000 emails per hour.
High-accuracy contact extraction requires structured schema enforcement and high FP8 throughput. The NVIDIA GB10 appliances featured on gb10.work provide unified high-bandwidth memory capable of serving Gemma 4 at sub-second latencies with zero power waste.
No. The extractor employs a smart non-destructive merge policy: it verifies existing directory entries, preserves existing notes, and enriches only missing attributes (such as newly discovered direct mobile numbers or updated job titles).