The systems that already run the business
Legacy ERPDesktop appsMainframesOld databasesCustom software
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. Its own symbolic and hybrid models turn measured state into explicit decisions, then operate your systems reliably without rebuilding anything.
macOS (universal) · Linux (x86_64 / aarch64) · checksum-verified · no runtime needed
Existing systems → modern business
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.
WorkflowsBusiness rulesDataVerified connections
Where your teams already get work done
Monday.comSalesforceServiceNowTeamsWorkday
Works with the enterprise software your business already uses.
Connect operations, sales, service, finance, HR, and collaboration tools while the systems underneath them keep running.
Monday.comSlackMicrosoft TeamsMicrosoft 365
Google WorkspaceSalesforceServiceNowHubSpot
WorkdaySAPOracleNetSuite
ZendeskSnowflakeOktaDocuSign
Proof in motion
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.
Live runtime proof
15.7 sec · direct capture
What Perslis is
Our models reason. Our runtime makes them operational.
Perslis builds its own symbolic models and hybrid models. They learn explicit rules, maintain observable state, and produce evidence-carrying decisions. The Perslis runtime gives those models memory, tools, perception, execution, and a verified operating loop.
Perslis models reason
Symbolic and hybrid models turn measured experience into explicit rules, spatial state, and decisions that can be inspected.
The loop →The runtime operates
It plans, executes, checks authority, recovers from failure, and records what actually happened—without handing control to a model that guesses.
Everything is a symbolic system
Your world already has rules. Perslis makes them operational.
Every organization is a system of entities, relationships, state, rules, and authority. Perslis maps that structure into a living symbolic model it can inspect, reason over, and operate. The domain changes; the architecture does not. Software, science, communication, robotics, and defense all become explicit systems—not black boxes.
Healthcare
Represent patients, evidence, clinical rules, permissions, and workflows explicitly—then act only when the required facts and authority are present.
Insurance
Turn policies, claims, exclusions, evidence, and approvals into readable relationships and rules that explain every outcome.
Finance
Map ledgers, controls, obligations, batch jobs, and mainframe state into a symbolic system that can be verified before it moves money.
Physical autonomy
Model rooms, objects, routes, hazards, goals, and authority locally—so machines can navigate and adapt without surrendering control to hidden weights.
COBOL execution proof
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.
The architect's loop
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"
01
Plan
Turns your outcome into a concrete, ordered approach.
02
Research
Maps the system, reads dependencies, and understands how it behaves.
03
Analyze
Weighs options, risks, and fits against the real project.
04
Decide
Judges if the code is right — verification first.
05
Build
Creates the change against the real system.
06
Test
Runs it and proves the result — checked, not assumed.
07
Deploy
Ships the verified outcome, and keeps it operational.
08
Connect
Wires legacy systems to modern API applications — the bridge from old software to your new stack.
Live Windows execution
23.3 sec · continuous loop
It learns you
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.
Symbolic AI
Perslis is symbolic at the point of decision.
Perslis models reason through explicit rules, facts, spatial state, and proofs a decision must satisfy. Hybrid models may use neural perception where useful, but symbolic authority decides what is allowed to act.
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.
Hybrid where useful
Neural perception can propose observations; Perslis symbolic models verify the critical decision.
Airtight by default
Your research never leaves your computer.
Perslis symbolic models and the runtime execute locally by default—your data, rules, discoveries, and reasoning stay on your machine. Cloud services are optional integrations, never the source of authority. You control everything.
Local Perslis models
Symbolic reasoning and rule execution run on your own hardware without requiring a cloud model to answer the question.
Never trained on your data
Nothing leaves your devices to train any model, ever. Your research and your discoveries are not someone else's training set.
You control everything
Optional external services never receive authority. Perslis rules, records, and decisions stay on your disk and keep working offline.
Dead-system reconstruction
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.
Live legacy workflow
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.
The moat is the runtime
The model and runtime are one operating system for decisions.
Perslis develops the reasoning models and the operating runtime together. Symbolic models hold authority; hybrid components can extend perception without turning execution into a black box.
Perslis symbolic + hybrid models explicit rules · spatial state · measured learning · optional neural perception
↓
Perslis Runtime
↓
MemoryPlanningVerificationSymbolicEvolutionDesktopTerminalLegacy systemsAPIs
↓
Reliable execution
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.
Field notes · Windows 95
Win95 USB Lab
From USB assumptions to testable evidence. Includes the complete white paper, 20 recorded host checks, and the plan for VM and physical testing.
Perslis projects
Jira Monkey
Jira intake, local Perslis triage, explicit routing rules, and a final verified review that returns proposals to the queue or to a human.
Fail-First Models
Failure becomes structure.
Our first model, Peel, learns only from failures it has actually had. Each failure becomes a rule that cites its evidence. Measured on Atari 2600, DOOM, Fallout and a drone course.
Research
White papers & projects
Full papers, experiments, and the reasoning behind each tool. USB Lab, Perslis, COBOL, Jira Monkey, and Liberate.