Technology · Artificial Intelligence
Multi-LLM, not a single model.
DEPLAW doesn't integrate AI as a single model, but as an orchestrated interplay of several language models — the right one for each task, and external, self-hosted or fully local depending on compliance needs.
01 · Which models
External, hosted, local.
DEPLAW doesn't commit to a single language model. Within one workflow, different models can handle different steps — orchestrated rather than delegated to a single LLM. Multiple models can also cross-validate each other to reduce errors and hallucinations before a result flows into the case file.
External
Anthropic, OpenAI, Google, Mistral and others — the highest model quality for general tasks.
Self-hosted
LLM resources we operate ourselves — control over infrastructure and model version.
Local
Run entirely within your own data center — for the highest compliance requirements.
02 · Integration
Every model can be connected.
Integration happens through a standardized API layer. New models — from Anthropic, OpenAI, Mistral or Google to a local LLM with a generic API — can be added at any time, without touching the workflow logic itself. This keeps DEPLAW independent of any single provider and lets it adopt new, more capable models as soon as they become available.
03 · Data protection
Pseudonymized, not just trusted.
Before a request reaches an external language model, its content can be pseudonymized: names, addresses and other identifiers are replaced with placeholders. The mapping key between placeholder and real person stays exclusively within our own infrastructure. The external model receives the request without real data; after the response comes back, the result is automatically merged back into the real case context.
Pseudonymization ≠ Anonymization
With anonymization, tracing back to the affected person is permanently excluded. With pseudonymization, it remains possible — but only via the separately stored mapping key. That's exactly what matters: only this way can the AI result ultimately be matched back correctly to the right case file and the right parties, instead of staying anonymous and therefore unusable for case handling.
For clients with especially high compliance requirements, we advise on the right choice of model and, where needed, set up dedicated, locally operated LLM resources — so data never leaves your own infrastructure.
Local LLM in the law firm: benefits & practical tips →04 · Usage
A lot of AI, always controllable.
At DEPLAW, AI isn't a side tool next to the process — it's embedded in many individual, clearly bounded process steps. Every use remains an auditable step with a defined input and output value — not an uncontrolled black box.
Workflow tasks
AI as a single, defined step in the BPMN process.
Legal AI Chat
Free-form questions about the case file — like a ChatGPT tailored to your own matter.
Inbox
Classification and summarization of incoming messages and documents.
Data extraction
Structured data from unstructured documents — see OCR & RAG.
Frequently asked questions
Artificial intelligence, explained briefly.
No. Before a request reaches an external language model, personal data is pseudonymized; the mapping key stays exclusively within our own infrastructure. The result is then automatically merged back into the real case context.
With anonymization, tracing back to the person is permanently excluded. With pseudonymization, it remains possible — but only via a separately stored key. That's exactly what matters: only this way can the AI result later be correctly matched back to the right case file and the right parties.
Yes. Which models are used for which task is configurable — from purely local models to a combination of several external providers with cross-validation.
No. Every AI use is a single, auditable workflow step with a defined input and output value, not an uncontrolled overall process.
Which AI configuration fits your compliance needs?
We'll work out together which models and which depth of pseudonymization match your requirements.
Schedule a technical conversation