AI as infrastructure, not another tool.
Law Flow helps law firms and legal departments become AI-native. We design the architecture, set it up, connect your documents and your know-how, and train your team. Confidentiality and data protection get settled at the design stage, before the first client file reaches the system.
Why AI has changed so little in law firms so far.
Expensive and locked in. Licence models tie the firm to one vendor, along with that vendor's roadmap and prices.
Knowledge scattered. It sits in mailboxes, in the DMS (document management system) and in individual heads, but not where the work happens.
No measurable effect. After the rollout, nobody can usually say how much time was actually saved.
What law firms tell us.
Three lines from intro calls, anonymised and cut to the point. They describe the situation our work starts from. Plus one voice from working together.
“The sheer number of tools is overwhelming. Just getting an overview costs so much already, and to go deeper than that there is often neither the time nor the energy.”
“In the confusing world of AI, Law Flow helped me separate what matters from the hype and showed me what is possible and sensible for my firm.”
“We get so many emails. Of course we deal with all of it, but sometimes something slips through.”
“What you really want is an AI that knows the fact pattern you have already talked about six times.”
This is what AI as infrastructure looks like.
AI as infrastructure. Manage context.
Build know-how into every step.
Secure foundation
Knowledge base
Context layer
You are not buying software. You get an environment that belongs to you.
No licence model
You are not renting access from us. Your firm contracts for model usage directly with AWS and pays AWS prices, with no margin on top and no fee per seat.
Not another tool to log into
No extra interface for you to paste case content into. What gets connected are the systems you already work in: mailbox, calendar, document storage, practice management software.
Not an AI agency
The advice comes from a fully qualified German lawyer who knows how firms run and what professional conduct rules require. You will not spend the first hour explaining why a statutory time limit is not a project deadline.
No lock-in to us
Account, configuration and documentation sit with the firm. You can end the engagement at any time and carry on with the same setup.
What you get out of it.
Time saved. An illustrative calculation: five hours saved per lawyer per week, at 250 € an hour across 48 working weeks, is roughly 60,000 € of additional capacity a year, provided the time freed up goes into client work.
Firm knowledge on call. Faster access to what the firm already knows, through a searchable knowledge base built from your own documents (RAG: before every answer, the model looks things up in your documents instead of writing from memory).
No shadow AI. Private accounts give way to use that is documented, auditable and governed by policy.
AI work as its own line item. We show you how to put AI work on the client's bill instead of letting it disappear into the hourly calculation.
Confidentiality holds. Your data sits in your own AWS account (Amazon Web Services, Amazon's cloud platform) in the EU, not in a vendor's account, and is not used for training. Professionals bound by legal secrecy also get a separate confidentiality agreement (§ 203 StGB and § 43e BRAO, the German rules on professional secrecy for lawyers).
It actually gets used. Rolled out with the whole team, not just the one colleague who likes AI: hands-on training, example cases from your own files, defined areas of use, instead of a licence nobody opens after three weeks.
Four building blocks. You start with one.
Each block can be commissioned on its own, and the starter package usually comes first. Whether and when the others follow is your call afterwards.
Starter package
The safe way in: your own AWS account plus a local AI workspace, set up to hold up under professional conduct rules.
- AWS confidentiality agreement for professionals bound by legal secrecy (§ 203 StGB, § 43e BRAO)
- Secure account setup with hardware MFA (sign-in requires a physical security key as well)
- Connection to Claude Cowork (the workspace for Claude, Anthropic's AI model)
- Acceptance testing and handover of the configuration
- Typical duration: 1–2 weeks
AI agents
Recurring tasks automated by purpose-built AI agents (programs that carry out several steps in sequence on their own).
- Needs analysis: “What eats the time?”
- Automation of your own workflows
- Test phase with fine-tuning
- Further development on request
Knowledge base
Making your firm's knowledge searchable and usable by agents.
- Review of the available data
- Building the knowledge base
- RAG implementation (looking things up in your own documents before every answer)
- Synchronisation and upkeep on request
Context layer
The live, situational context the agent works from at runtime, alongside the knowledge base.
- Review of the available context sources
- Access and confidentiality rules
- Connection to calendar, email and practice management software
- Upkeep and fine-tuning on request
A licence from a vendor, or your own account.
Both routes can produce AI use that holds up under professional conduct rules. They differ in who owns what was built once the work is done.
| Criterion | Licence from a vendor | Your own account |
|---|---|---|
| Contract partner | The vendor that resells the model | AWS, directly with your firm |
| Price per model call | List price plus a margin, up to 20 % with comparable offerings | AWS list price, no middleman |
| Running costs | A fee per seat, whether or not it gets used | Billed by actual consumption, no seat fee |
| Choice of model | Whatever the vendor has integrated, often late | Any model enabled in your region |
| Confidentiality | Whatever the vendor’s terms and conditions say | A separate agreement for professionals bound by legal secrecy (§ 203 StGB, § 43e BRAO) |
| Logging | Whatever the vendor shows in its own interface | Complete logs in your account, reportable per lawyer |
| Your own workflows | Live inside the vendor’s interface | Sit as files in your environment, readable and versionable |
| If you switch | Access ends and what you built stays with the vendor | Account, configuration and documentation stay with you |
The comparison describes the typical case, not every single offering on the market. Which one adds up for your firm depends mostly on the number of lawyers and on how heavily the system actually gets used. Two figures are enough for the order of magnitude, and we work them through in the intro call.
Not sure which block comes first? We settle that in 30 minutes.
Book an intro callThe foundation: defensible under professional conduct rules and free of vendor lock-in.
Why AWS Bedrock
What Bedrock (the AWS service that provides language models) brings to use that has to stand up professionally:
- Encryption in transit and at rest.
- Hosting in Germany or the EU, enforced technically through Service Control Policies (account-wide blocks that not even an administrator can lift).
- Access control (IAM) and logging (CloudTrail), with certifications including ISO/IEC 42001 (management systems for AI), ISO 27001 (information security) and SOC (externally audited provider controls).
- AWS confidentiality agreement for professionals bound by legal secrecy (§ 203 StGB, § 43e BRAO).
- Bought direct, with no reseller margin: you pay AWS prices per token (the billing unit for processed text). Comparable BRAO-compliant offerings add margins of up to 20 %.
Why Law Flow
- Advice from a fully qualified German lawyer with a technical focus, not a generic AI agency.
- An eye for professional conduct rules and how your working day actually runs.
- You keep all of it: the architecture, the know-how and your independence from tools and models.
- Demonstrable: in the intro call we show an anonymised client setup running, not just slides.
- Verifiable: our open-source library for law firm workflows is public: legal-ops-germany on GitHub.
Notes on AI-native legal work.
Agentic AI in legal work: what the numbers show
Adoption, cancellation rates, the end of the hype: what Thomson Reuters, Gartner and Forrester say about agentic AI, and what follows from it for law firms.
ReadLaw Flow OS: eight AI agents and a coordinator that barely thinks
How I run Law Flow with eight AI agents on Claude Code: an event bus, a LangGraph coordinator on AWS in Frankfurt and approvals that no agent can grant.
ReadMaking firm knowledge accessible to AI: RAG is not the only answer
RAG splits documents into chunks and pulls out the most similar ones. In file work, the most similar passage is rarely the right one.
Read
Law and technology from the same person.
Elias Reiche is a fully qualified German lawyer (Volljurist), a former research associate in IP and IT law, a legal tech founder and a student of legal informatics. That combination of legal training and technical focus shapes the advice.
With Law Flow he helps firms not just introduce AI but make it compliant with professional conduct rules and genuinely usable day to day. No vendor pitch, just implementation that works on real matters. Law Flow is deliberately a one-person practice: you always speak directly to the person who builds it.
Let's talk about the biggest lever in your firm.
A free intro call works out where it is. The first ten minutes are yours: how you work today, what is actually missing, how many lawyers are affected, who decides and by when. Then we show an anonymised setup running and say which block makes sense first.
Afterwards you get it in writing: a one-page assessment with the recommended first block, the price for it and the points to settle beforehand. You can pass it on internally, even if you decide against us.
Book an intro callLaw Flow is a one-person practice. You speak to the person who then carries out the setup himself. So we take on only as many firms as can be looked after properly that way.
If you have a fixed date by which something has to be running, say so at the start. Then both of us know within the thirty minutes whether it fits. If the calendar is full, we say so in the call instead of moving the date.
// 30 minutes, free, by video.