The AI operational runtime
Every company has software
nobody understands anymore.
Perslis fixes that. It's an AI operational runtime that understands, maps, connects, and runs your existing software systems — the layer that lets AI workforces (LangGraph, MCP, CrewAI, or your own agents) operate your systems reliably, without rebuilding anything.
macOS (universal) · Linux (x86_64 / aarch64) · checksum-verified · no runtime needed
Upgrade your infrastructure without replacing what works.
Perslis maps how your existing software, data, and workflows actually run, then connects them to the enterprise platforms your teams already use — without a risky rewrite or a migration project that takes years.
Understand. Map. Connect. Keep running.
Where your teams already get work done
Watch Perslis bridge code, Windows 95, and DOS in one live run.
This is operating proof, not a concept animation. The 15-second capture shows one runtime coordinating source, a Windows 95 guest, and DOS execution while keeping the evidence visible.
A shell that powers AI — not another model
Perslis has no model of its own — no AI to sell you. It is the body: the harness, shell, and loop that any model plugs into. Bring OpenAI, Anthropic, Google, Qwen, DeepSeek, or a local model — Perslis is model-agnostic, and it uses all models, for all jobs.
Models think
Use any brilliant reasoning engine you choose — frontier or local, whichever fits the task.
The loop →Perslis operates
It plans, routes each step, recovers from failure, and ensures the work is actually done — a coordinated, reliable process, not one model guessing.
Perslis is whatever you need it to be.
Perslis adapts to your industry and your systems. It learns the domain, connects the AI, and operates it — one universal runtime, shaped to your world.
Healthcare
Ingest EMRs and clinical systems, wire them into modern AI, and keep patient workflows running — safely and verifiably.
Insurance
Operate claims engines and underwriting logic nobody fully understands, and connect AI into them without a rebuild.
Finance
Run the batch jobs, ledgers, and mainframe logic the business depends on — mapped, verified, and modernized in place.
Anything
Manufacturing, government, logistics, your own stack — if it's software, Perslis can understand, operate, and maintain it.
A legacy language becomes a live, playable system.
COBOL Racer turns a batch-era language into a terminal racing game, proving the same runtime can compile, launch, and operate unconventional software.
You don't write code. You direct the model — it executes.
Describe the outcome in plain language. You direct; the model, inside Perslis's body, executes — it plans, researches how the system actually works, analyzes the options, decides whether the code is right, then builds, tests, deploys, and connects it to your modern APIs. You make the calls; the runtime does the work and proves it. Connecting legacy systems to modern API applications is the point.
$ perslis "get the nightly ETL job talking to the new warehouse and prove the row counts match"
Plan
Turns your outcome into a concrete, ordered approach.
Research
Maps the system, reads dependencies, and understands how it behaves.
Analyze
Weighs options, risks, and fits against the real project.
Decide
Judges if the code is right — verification first.
Build
Creates the change against the real system.
Test
Runs it and proves the result — checked, not assumed.
Deploy
Ships the verified outcome, and keeps it operational.
Connect
Wires legacy systems to modern API applications — the bridge from old software to your new stack.
No two Perslises are alike.
The more you use Perslis, the more it adapts to your systems, standards, preferences, and personality.
Persistent memory
Perslis remembers your decisions and conventions across sessions.
Self-evolution
It tunes how it plans, routes, and verifies — getting better at your work specifically.
Shaped personality
Tone, defaults, and judgment thresholds adapt to how you operate.
A model can be confident and wrong. Symbols can't.
Perslis runs a symbolic layer alongside the model — hard rules, facts, and proofs a decision must satisfy. If the model proposes something that contradicts it, Perslis rejects it.
Rules & facts
Symbolic invariants and constraints make decisions checkable logic, not vibes.
Grounded verdicts
Decisions are backed by evidence and proof, not a model's say-so.
Neural + symbolic
Language models handle the open-ended part; symbolic logic verifies the critical part.
The 30-year-old Windows app that still runs the business.
Hand Perslis a legacy binary — .exe, .com, DOS or Win32,
16- or 32-bit. It runs natively on a modern Mac — no Windows, no Wine, no
emulator — and keeps running as everything around it changes.
Inspect, inject, and verify a Windows-era target.
Terminal evidence and guest state stay visible side by side, so an operator can see what changed and whether the target actually ran.
We don't compete with models. We orchestrate them.
Perslis ensures whichever AI model is best for the job can actually get work done reliably — without changing your underlying architecture. A better model tomorrow plugs right in.
Start mapping your trees today.
Point Perslis at a system and watch it build the map — dependencies, windows, workflows, the whole shape of what you've got. One line to begin.