Deployment
Telha is designed to run inside the customer environment, connect to approved business systems and keep investigation data under customer control.
Evidence, prompts and AI output stay inside an environment you own, whichever option you choose.
Telha and its AI models deploy into an Azure, AWS or Google Cloud subscription you own. Your existing cloud agreement, regions and audit controls apply unchanged.
For restricted networks and strict residency mandates, Telha runs on a single GPU server on your premises. The model sits beside the evidence. Nothing needs to leave the building.
Three capacity profiles cover most investigation teams. Same models, same features: they differ only in responsiveness and how much runs in parallel.
| On-demandAI compute starts when a case needs it, shuts down after. | StandingMost teamsOne dedicated AI instance, ready throughout the working day. | High-throughputDedicated capacity for parallel casework and archive-wide analysis. | |
|---|---|---|---|
| Best for | Periodic investigations, a few cases a quarter | A standing team with weekly casework | Multiple teams or heavy consolidation workloads |
| First response after idle | Under ~2 minutes while compute spins up | Immediate | Immediate |
| Parallel workload | One analysis at a time | A small team working concurrently | Concurrent teams, plus scheduled archive-wide jobs |
| AI hardware held | None, billed per second of use | One 24 GB-class GPU | One 48 GB-class GPU, or a pair |
| Indicative compute cost | From ≈US$50 / month | ≈US$600 to 1,400 / month | ≈US$1,400 to 2,700 / month |
| Available on | Cloud tenancy | Cloud or data centre | Cloud or data centre |
Indicative figures at typical mid-2026 cloud GPU rates, billed by your cloud provider at cost. Telha does not resell compute. On-premises hardware is a one-time purchase with no per-token fees. Exact sizing is confirmed during an architecture review.
Start with a bounded investigation workflow and expand source coverage deliberately.
Install Telha in the agreed customer-controlled environment.
Map enterprise users, roles and investigation approvers.
Connect the systems required for the initial investigation type.
Define mandate, scope, retention and evidence rules.
Validate the complete workflow with representative evidence.
Telha deploys an open-weight model inside your environment, so investigators get AI assistance without evidence ever reaching an external service.
Deployed with Telha and sized to your hardware. You can inspect the model, pin its version and upgrade on your own schedule.
Summarise threads, draft chronologies and surface gaps, computed only over the evidence admitted to the case.
Prompts, embeddings and evidence stay on your infrastructure. No external AI APIs, and no vendor training on your data.
Telha does not require an immediate replacement of Microsoft Purview, archives, forensic tools or existing review systems.
Keep Microsoft-native preservation and legal-hold processes where they are already established.
Bring approved exports and forensic collections into the same governed case model.
Connect evidence across systems and turn it into chronology, findings and defensible output.
Architecture workshop
Review sources, identity, network boundaries, deployment, evidence retention and operational ownership.
