Describe the process. Get a workflow that works.
InterlaceLogic turns plain-language requests into working business workflows that fetch data, apply your rules, ask the right person to approve and send the result. Every draft is planned with you, built step by step, and tried on real data before anything is saved.
- Runs on your servers
- Your choice of AI model
- People approve before anything is sent
- Fetch the vendors
- Work out days to expiry
- Flag and give a reason
- Build the report
- Ask you to approve
- Email finance
From a sentence to a workflow you can trust
Most AI builders try to write the whole workflow in one go and hope it runs. The Creator works the way a careful colleague would: it agrees the plan with you, builds it piece by piece, then proves it works.
Describe
Explain the process in your own words: where the data comes from, the rules, who approves, what gets sent.
Asks when something's missingConfirm the plan
A short plan comes back: the steps, the checks it will make, and the rules you stated, like "no AI" or "approve before sending".
You decide before it buildsBuilt and checked
Each step is built on its own and validated. Your rules are enforced, and anything that breaks one is repaired.
Small steps, no timeoutsTried on real data
The draft runs on your real data with emails, files and approvals simulated. The checks run on the results, and failures are fixed before you see the draft.
See exactly what people will getSee the result before anyone else does
Before a draft reaches you, it runs on your real data with everything that leaves the system simulated. You review the approval request, the email and the report exactly as they will look, and nothing is sent or saved.
- Real reads, simulated sends. API reads, files, calculations and AI steps run for real. Emails, saves and approvals don't.
- Problems fixed automatically. A wrong field name, a blank date or text that reads like code goes back to the builder with the data that caused it.
- Changes are checked too. Ask for an edit and the changed draft goes through the same checks again.
| Vendor | Contract expiry | Reason |
|---|---|---|
| Northgrid Applications | 2026-11-12 | Expires in 42 days |
| Bytecairn Systems | 2026-06-28 | Contract expired 95 days ago |
| Stratosphere Hosting | missing | Not approved (Pending Review); missing contract expiry |
| Clearview HVAC | 2027-06-30 | Not approved (Suspended) |
Your intent, written down and enforced
The plan's checks become tests on the real results, such as "every flagged vendor has a name" or "approved vendors are never called unapproved". Your hard rules apply to every version of the draft.
- Tests tied to the data. Checks can apply only where they make sense, for example "for rows where approved is true".
- Rules that can't be quietly broken. No AI steps, only these connections, a person approves before anything is sent.
- Honest results. If something can't be fixed, the draft says so, points to the step, and keeps the last version that worked.
Trial run: it ran, but 1 of 8 checks failed:
Every flagged vendor has a reason (item 3 is blank)
Trial fix 1: reason now covers vendors without a date.
Trial run: completed with real data in 0.2 s; 8 of 8 checks passed.
Everything a business workflow needs, on one canvas
Build visually or with AI, then run, schedule and audit. Over 40 built-in steps, plus your own.
Visual workflow editor
Connect steps on a canvas with branches, loops and merges. Validation as you work, versions on every save, live progress on every run.
AI where it helps
Summarise, extract, classify and translate with the model you choose. Deterministic steps handle the maths, dates and rules, so figures never depend on an AI's guess.
People in the loop
Approvals, checklist reviews and forms pause a run until the right person responds. Every decision is recorded in the run history.
Connections, not keys
Administrators define the APIs a workspace may call. Workflows refer to a connection by name, so credentials never appear in a workflow.
Email that stays in bounds
Admin-defined mail servers with a fixed sender, allowed recipient domains and hourly limits. Attach the reports a workflow creates.
Documents in and out
Read PDFs, Word and Excel files and spreadsheets from workspace folders. Produce clean PDF, Word and Excel reports with tables that fit the page.
Data you can trust
Filter, select, combine, tabulate and calculate with visible workings, date arithmetic included. Each step's output can be inspected in the run log.
Python when you need it
Drop a Python step into any workflow. Scripts run in a sandbox with no network access and strict limits.
Runs that survive restarts
Every completed step is saved as it happens. After a restart, an unfinished run carries on where it stopped. Completed steps aren't repeated, so no email goes out twice.
Built for teams that answer to an auditor
AI suggests; people and policies decide. InterlaceLogic is designed so that the convenient path is also the safe one.
Self-hosted
Runs on your own infrastructure. Workflows, files and run history stay in your environment.
Secrets stay with admins
Connection credentials are encrypted at rest and added by the server at call time. They are never shown back or stored in a workflow.
Workspaces and roles
Personal and shared workspaces with Owner, Editor and Viewer roles. Administrators manage users, models and integrations.
Guarded network access
Outbound requests go only to approved connections and public addresses, with each connection's methods and paths restricted.
AI activity you can see
Every model call is logged with its size, timing and outcome. Capturing prompt text is off by default and controlled by administrators.
Nothing runs by surprise
AI drafts are never run automatically. Trial runs can't send, save or ask anyone, and steps the AI couldn't build are clearly marked "Needs implementation".
Workflows teams build in an afternoon
Each of these started as a few sentences in the WorkFlow Creator.
Vendor compliance check
Flag vendors whose contracts have expired or end within 90 days, or who aren't approved. Give each a plain reason, have a finance lead approve the report, then email it.
Onboarding gap report
Pull every employee from the HR system, keep those who've been hired but haven't started, and produce a PDF for the people team with start dates and managers.
Invoice intake triage
Read new invoices from a folder, extract the totals, check them against approval limits, and route anything over the limit to a person with the reason attached.
Contract review assistant
Loop over contract documents, have AI pull out renewal dates and notice periods, calculate the deadlines, and send a weekly digest of what's coming up.
Bring your own model. Add your own steps.
Connect any OpenAI-compatible endpoint and switch between profiles, such as development and production, from the admin console. Extend the catalog with .NET plug-ins or Python extensions; new steps appear in the editor and the Creator without changing any prompts.
- Several model profiles, one active at a time, with the request settings each server needs.
- Public SDK for custom steps, loaded when the server starts.
protected override NodeManifest Describe() =>
NodeManifest.Create("acme.greet", "Greet")
.Category("Acme")
.Input("name", DataTypes.Text, required: true)
.Output("greeting", DataTypes.Text);
Questions teams ask first
Does the AI run my workflows?
No. The AI drafts workflows; it never runs them on its own. Workflows run when a person starts them or on a schedule you set. AI steps inside a workflow are explicit and visible on the canvas, and you can forbid them for a workflow.
What does a trial run actually do?
It runs the draft once on your real data so problems show up before you save it. Steps that read or calculate run for real. Sending email, saving files, changing data in other systems and asking people for approval are all simulated. You see exactly what would have been sent, and nothing is.
Which AI models can I use?
Any OpenAI-compatible endpoint: hosted services such as OpenAI and Azure OpenAI, or models you run yourself with Ollama, vLLM or LM Studio. Administrators set up profiles, choose which models people may use, and switch between them.
Where is my data kept?
On the servers where you install InterlaceLogic. Workflows, files, run history and settings are stored there. Data only goes to an AI model when a workflow step or the Creator calls the model you configured.
Can it connect to our internal systems?
Yes. Administrators define connections to your APIs once, with the credentials, allowed methods and allowed paths, and workflows use them by name. For anything else there are Python steps and the plug-in SDK.
What happens if the server restarts during a run?
The run continues when the server comes back. Completed steps are restored from a journal rather than repeated, and an approval that was pending is requested again.
See your process built in a single conversation
Bring a process your team runs by hand today. We'll describe it to the Creator together and you'll see the trial run results.