OPEN MODULE STANDARD v0.1
A module is a small folder of plain files.
The standard is free and public. A module teaches PathWise to troubleshoot one thing: a vendor, a tool, or a layer of the access path. The engine already knows how to reason, cite evidence, refuse to make changes, and return an answer for a person to act on. A module only supplies the know-how and where to read the data.
This is a first public draft. The parts marked Shipped run today. The parts marked Designed are specified so you can build for them, and they ship in a later release.
What a module supplies
How you, the expert, check this tool and tell healthy from broken. You write it in plain language. If you can explain your troubleshooting to a new hire, you can write this file.
A place the engine reads evidence from: a file you export and drop in a folder, a vendor management API, or a vendor MCP server. The engine reads it. It never writes back.
The engine owns the rest. Safety handling and evidence checking live in the engine, not the module. A module author cannot make the tool unsafe, by design. A module is just files, so you share it as a folder. No account, no upload, no gatekeeper.
What a module contains
module-name/ module.yml required who and what this module is (the manifest) knowledge.md required how you troubleshoot this, in plain language connection.yml optional where to read the data (defaults to file drop) evidence.yml optional field labels, for a later feature (not read yet) examples/ optional sample data and a sample question README.md optional extra notes for other authors
The engine validates the manifest against its schema when it loads, and rejects an unknown or misspelled layer. It checks that module.yml and knowledge.md exist and are not empty. The optional connection.yml is read if present and checked for basic shape. Only a file-drop connection is accepted today. evidence.yml is not read by the engine yet.
The knowledge file
This is the part only you can write. It is plain markdown, no code. The engine reasons from it.
# What this module checks One or two sentences. # What healthy looks like - Bullet list of good states. # What broken looks like - Bullet list of failure modes. # What to check, in order 1. Ordered troubleshooting steps. # Common causes and how to spot them - Cause: X. Spot it: evidence pattern Y.
Four doors to your data
Pick the door by your comfort level. Everyone can build a module. Nobody is forced to code. The engine tries them in a fixed order: MCP, then the API, then the file drop. A module whose live door is not configured still loads and still teaches, and reads nothing until you configure it.
File drop
Export data from your tool to a JSON or CSV file and drop it in the module folder. No code, no authentication. The engine reads the file and never writes to it. This is the default when a module declares nothing else, and eight of our nine modules use it.
Vendor API
A read-only connection to a vendor management API. The module declares the base URL and the name of a credential, never the credential itself. The operator stores the real value in Settings, encrypted, and the engine reveals it only at the moment of the outbound call.
MCP server
A read-only connection to a vendor MCP server. The module must list the exact read tools it uses, and a module that omits that list fails to load. The engine never hands the model a vendor server's full tool surface.
Custom adapter
A developer escape hatch: a small adapter that returns rows the engine reads. The signature is defined here. The engine does not accept it yet, and a module that declares one fails to load rather than loading half-configured.
The eight layers a module can read
The access path is vendor-neutral. A module declares the one layer it reads. Vendors are modules, never layers.
- deviceEndpoint health, config, posture
- local networkLAN, Wi-Fi, gateway, VPN tunnel
- userAuthentication and identity
- rightsAuthorization, firewall and access policy
- resolutionName to address: cache, hosts, DNS
- routingReachability and the listening port
- session-trustTLS handshake, certificate, cipher, clock
- destinationThe service is listening, healthy, backend up
What the engine guarantees
- No changes. A module's interface has one method, a read. There is no write, execute, or mutate method to implement, and the engine stamps every module tool read-only itself. That holds on a live API or MCP connection exactly as it holds on a dropped file. A module author does not get to set that field.
- Evidence, not opinion. The engine grounds each claim in the data the module supplied, with a confidence level and a note on what it could not verify.
- Validation. The manifest is checked against a schema at load time, and an unknown layer is rejected. Required files must be present and non-empty.
- Open standards. Where a standard already exists, the format uses it, so your module is portable and not locked to us.
What v0.1 ships today
- Shipped Plain knowledge files, manifest validation, the eight-layer model, three source types (file drop, vendor API, MCP server) with a fixed resolution order, and nine modules including DNS and Check Point. Eight read through the file drop. Check Point declares a Management API door, tested against fixtures rather than real hardware.
- Designed The custom adapter door, and schema validation of the optional connection and evidence files.
See the nine modules that ship today.
Who is it for?
Anyone with the right permissions stands it up. After that, anyone working an access ticket, help desk, IT ops, or developers, gets the same evidence-backed answer.
What exactly do I get?
The engine skeleton as a Docker image, all nine modules, the command-line tool, and the browser interface. You supply your own model backend.
Does it change anything in my network?
No. It diagnoses and advises. It runs checks, including its own live network probes, and proposes a fix you perform. It makes no changes.
It is built with AI. Why should I trust it?
PathWise is built by an engineer with more than 30 years in systems infrastructure, 18 of them securing networks in regulated banking, holding the CISSP and CCSP. It solves a problem its builder has lived for two decades. You still do not have to take it on faith. PathWise cannot change your systems. The read-only boundary lives in the code, not in a request to the model, so the worst case is a wrong answer you catch, never a wrong action. Every answer shows the probes it ran and the evidence behind each claim, so you check the conclusion instead of trusting it. It ships with more than 1,600 automated tests, and an outside security review of the live app in July 2026 found no critical or high issues. And it is free to trial on your own box until the beta license expires in one year. Point it at a problem you already understand and watch what it does before you pay.
What model does it run on?
You bring your own. It runs against the Anthropic API by default, or any OpenAI-compatible backend, including a local model through Ollama or vLLM. A managed in-tenant placement is planned. The model is a quality dial you set.
Is my data sent anywhere?
Your model does the reasoning, so in the Anthropic mode your investigation data goes to your AI provider over your own key. Run a local model to keep the reasoning inside your boundary. A managed in-tenant cloud placement is planned. Probes send only the host or IP being checked.
What is a module?
A small folder of plain files describing one vendor or one layer. The standard is free and open, and you can write your own.
How is it licensed?
An annual subscription license. Solo covers one engineer. Team is licensed to one organization and covers four user accounts, with more sold in 3-packs. Modify it for internal use, no resale, as-is, copyright retained.