Pick one job
Start with one repeatable problem: documents, client intake, support tickets, internal knowledge, code review, or another process that eats staff time.
Coming soon. LLM Foundry is being prepared for early Fit Reviews. To register interest, email info@llm-foundry.com.
How it works
A private AI process your team can understand.
Foundry sits beside your existing tools, helps with the first pass, keeps a record, and queues important outputs for your team to check.
1pick one job
2agree what stays controlled
3set up the office system
4teach the steps
5use the right tool
6keep a record
7your team checks
8improve over time
The simple version.
Agree what must stay controlled
We look at the data, who should see it, where it currently lives, and which steps must be checked by a person.
Set up the office system
Foundry runs on Apple hardware controlled by your business. We help decide whether your current hardware is suitable or what you would need.
Teach Foundry the steps
We map what should happen first, second, and third: read this, extract that, search this folder, draft this note, flag this exception, stop here for review.
Use the right tool for each step
Simple jobs get a fast tool. Document jobs get document tools. Careful summaries and drafts get a stronger AI step. The buyer does not need to manage this; Foundry is configured around the workflow.
Keep a record
Foundry can keep a practical trail of what arrived, what it read, what it produced, what it flagged, and what is waiting for approval.
Your team checks the important outputs
Foundry prepares the work. A person approves client messages, legal or compliance-sensitive outputs, document results, escalations, and important decisions.
Improve it over time
As better tools become available, parts of the setup can be upgraded without starting the whole process again.
The tools stay familiar. The sensitive AI work becomes easier to control.
Work comes in
- Email attachments
- Matter packs
- Shared folders
- Helpdesk tickets
- Internal documents
- Code changes
Foundry helps with the first pass
- Reads
- Sorts
- Extracts
- Searches
- Drafts
- Flags
- Records
Your team decides
- Checks the source
- Reviews missing items
- Approves drafts
- Handles exceptions
- Sends back poor outputs
- Takes the next step
For agreed workflows, the AI work can run inside your controlled business environment. Your existing cloud tools may still be used for normal business tasks.
Right tool, right step
Not every job needs the same kind of AI.
A scanned document, a support ticket, an internal search, and a careful client reply are different jobs. Foundry can use different local tools for different parts of the process.
| Part of the job | Plain-English tool | Why it helps |
|---|---|---|
| Sort incoming work | A fast sorter | Keeps the queue moving |
| Read scanned pages | A document-reading tool | Helps capture fields from files |
| Search internal sources | A source search tool | Keeps answers tied to your own material |
| Draft a careful summary | A stronger AI drafting step | Gives your team a better first draft to review |
| Approve the result | Your team | Keeps judgement with the business |
Foundry is useful because it fits the tool to the job, then stops important outputs for review.
Foundry does not replace your business systems.
Can move into the private setup for agreed jobs
- AI help on selected sensitive documents
- Ticket reading and response drafting for agreed queues
- Internal knowledge search over selected private documents
- First-pass code review
- Work sorting and exception flagging
- Review queues and records for agreed jobs
Usually stays where it is
- Email delivery
- Case management, CRM, or accounting systems
- Helpdesk platforms
- GitHub, GitLab, or Bitbucket hosting
- Cloud backups
- Video calls and collaboration tools
- External research
- Customer-facing systems of record
Move the sensitive AI work where it makes sense. Do not rebuild everything that already works.
From fit check to first workflow.
| Phase | What happens | Typical time |
|---|---|---|
| Fit Review | We review the job, data sensitivity, volume, tools, approval needs, current AI spend, and supplier concerns. | 30–60 minutes |
| Recommendation | You get a practical yes/no and the best first job to test. | After review |
| Boundary design | We define what moves into the private setup, what stays cloud, what must be recorded, and where approval steps sit. | Scope-dependent |
| Setup | Foundry is installed and configured on suitable Apple hardware. | 1–2 days for supported setups |
| Testing | Sample documents, tickets, queries, or code changes run through the workflow with review checks. | Included in setup |
| Handover | Your team learns the review queue, visible records, exceptions, and support route. | 1–2 hours |
| Live | Foundry runs the agreed job with approval steps for important outputs. | Ongoing |
Timelines depend on access, hardware readiness, workflow complexity, data quality, security requirements, and integration scope.
What should be visible.
- What came in.
- Which source was used.
- What Foundry produced.
- What was flagged.
- What needs approval.
- Who checked it.
- What happened next.
This record can help with internal checks and governance. It does not replace legal advice, data-protection work, contracts, policies, security controls, or professional judgement.