Skip to content

vitron-ai/alethia

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
LICENSE
LICENSE
 
 
README.md
README.md
 
 

Repository files navigation

Alethia

The patent-pending zero-IPC E2E test runtime built for AI agents.

45× faster than Playwright on the localhost test loop. Fail-closed by default. Cryptographically chained audit packs. Local-first. Zero telemetry by default. Opt-in cloud. No CDP.

npm License: MIT Patent Pending Status

Why Alethia exists

Your AI agent generates a Next.js app on localhost:3000 and your users want it verified. Playwright adds 600ms of CDP marshalling tax to every assertion. Worse, it's async-by-construction — the agent's decide-act-verify loop is racing against stale DOM snapshots between every await.

Alethia is a different architectural shape. The driver and the DOM live in the same V8 isolate. Reading document.querySelector('button').textContent is a function call, not a network round-trip.

Playwright (CDP) Alethia (zero-IPC)
Avg latency per step 580 ms 13 ms
p95 latency per step 654 ms 24 ms
Process boundary 3 (test ↔ driver ↔ browser) 0 between driver and DOM
DOM access async, marshalled, race-prone synchronous, in-process
Per-step safety policy none VITRON-EA1 fail-closed gate
Audit trail trace viewer (debugging) SHA-256 chained, Ed25519 signable
Telemetry on by default in cloud product off by default; opt-in only
Patent moat none U.S. App 19/571,437

Benchmark: click-assert-wait scenario, 20 iterations, full numbers in the evidence pack.

Status

Alethia is in design-partner alpha (v0.2.0) as of April 2026.

  • ✅ MCP bridge published to npm: @vitronai/alethia
  • ✅ MCP bridge source open and auditable: vitron-ai/alethia-mcp — MIT
  • ✅ Patent filed: U.S. Patent Application No. 19/571,437 (non-provisional, claiming priority to 63/785,814 filed April 9, 2025)
  • ⏳ Public binary releases of the desktop runtime — coming with the design partner wave
  • ⏳ Headless mode for unattended CI agents — v0.3 milestone
  • ⏳ Cloud dashboard (signed evidence as a service, team collaboration) — post-MVP

If you build AI coding agents and want early access to the desktop runtime, email gatekeeper@vitron.ai. Design partner program is open and free during the alpha.

How it works

┌────────────────────────┐
│  Your AI agent         │  Claude Code · Cursor · Cline · Continue · ...
│  speaks MCP stdio      │
└──────────┬─────────────┘
           │ stdio (newline-delimited JSON-RPC)
           ↓
┌────────────────────────┐
│  @vitronai/alethia     │  npm package, ~9 KB, MIT, source on github
│  stdio → HTTP shim     │  Zero telemetry. Loopback only (enforced).
└──────────┬─────────────┘
           │ HTTP POST 127.0.0.1:47432 (loopback only, never networked)
           ↓
┌────────────────────────┐
│  Alethia desktop app   │  Electron main process — proprietary, patent pending
│  local JSON-RPC server │  Loopback bind, never reachable from network
└──────────┬─────────────┘
           │ webContents.executeJavaScript('window.__alethia.tell(...)')
           ↓
┌────────────────────────┐
│  Alethia renderer      │  Electron renderer — IS the browser
│  zero-IPC runtime      │  tell() → NLP compiler → Action IR
└──────────┬─────────────┘  VITRON-EA1 policy gate (per-step, fail-closed)
           │ direct synchronous DOM access
           ↓
┌────────────────────────┐
│  The page under test   │  Your localhost dev server, your file://, your app://
└────────────────────────┘

Two process boundaries between your agent and the runtime (agent ↔ shim, shim ↔ Electron). Then zero boundaries between the runtime and the DOM. That's the architectural difference that makes Alethia 45× faster than Playwright on the localhost test loop.

Try it

⚠️ Design-partner alpha. The desktop runtime is required and is currently distributed by request. Email gatekeeper@vitron.ai for early access. Public binary releases are coming with the v0.3 milestone.

1. Install the MCP bridge from npm (live now)

npm install -g @vitronai/alethia

2. Verify install

alethia-mcp --version
alethia-mcp --help
alethia-mcp --health-check

The health check will tell you the desktop runtime isn't reachable yet — that's expected if you don't have the binary. Once you do, it'll print:

✓ Connected. 5 MCP tools available.
  runtime version:  0.1.0-alpha.1
  default profile:  controlled-web
  kill switch:      inactive

3. Configure your agent

Claude Code (~/.claude/mcp.json)

{
  "mcpServers": {
    "alethia": { "command": "alethia-mcp" }
  }
}

Cursor (Settings → MCP → Add server)

{ "alethia": { "command": "alethia-mcp" } }

Cline / Continue / any MCP-compatible client

Same shape. They all speak the standard stdio MCP protocol.

4. Drive Alethia from your agent

"Use alethia_tell to navigate to localhost:3000, sign in as admin@example.com / password123, and verify the dashboard heading is visible."

The agent will call alethia_tell with that NLP. Alethia compiles it to Action IR, runs it through the VITRON-EA1 policy gate (default: controlled-web, fail-closed), executes step-by-step with synchronous DOM access, and returns a signed PlanRun with per-step results, policy audit records, and a SHA-256 integrity hash.

The 5 MCP tools

Tool Purpose
alethia_tell Execute natural-language test instructions. The headline tool. ~13 ms per step.
alethia_compile Compile NL → Action IR without executing. Preview before you run.
alethia_status Health + identity probe. Version, profile, kill switch state, driver stats.
alethia_activate_kill_switch Halt all automation immediately. Optional reason logged in audit trail.
alethia_reset_kill_switch Clear an active kill switch. Re-enables tell() calls.

Full docs: npmjs.com/package/@vitronai/alethia

VITRON-EA1: Ethical Automation v1

The first browser automation framework with built-in runtime-enforced ethical guardrails. Every step passes through a fail-closed policy gate before execution and produces a cryptographically chained audit record. The gate is in the same process as the executor — agents cannot bypass it.

Default classifications:

  • read (navigation, scraping, assertions) — always allowed
  • write-low (form input, drafts) — allowed by default
  • write-high (delete, purchase, transfer, submit-payment) — blocked by default
  • blocked (always denied)

The NLP compiler infers intent: "delete the user account" → !://write-high CLICK :text(user account). "type my password" → !://write-high TYPE blocked unless explicit allowSensitiveInput: true.

VITRON-EA1 is positioned as a publishable standard — other automation runtimes will be able to claim conformance via Ed25519-signed audit records verified against a public key registry. The conformance scheme binds adoption to the patent license: registered conforming implementations must implement per-step in-process enforcement, which is the patented method.

Privacy & security

Alethia is local-first with zero telemetry by default. Some of these guarantees are architectural (enforced in code, not policy); others are policy commitments that hold in v0.2 and any future opt-in cloud features will be clearly disclosed.

Architectural guarantees (enforced in code, can't drift):

  • Loopback only. The MCP bridge only speaks to 127.0.0.1. The desktop runtime's production webRequest filter blocks all non-file://, non-app://, non-localhost requests at the network layer.
  • Zero IPC. No ipcMain/ipcRenderer/contextBridge/preload — enforced by a CI gate (scripts/check-zero-ipc.sh) that fails the build if any forbidden API appears.
  • Sandboxed renderer. sandbox: true, contextIsolation: true, nodeIntegration: false, webSecurity: true.
  • Auditable bridge. The MIT-licensed npm bridge is ~530 lines. Read it in 5 minutes and verify it does nothing but forward MCP calls to localhost. Source: github.com/vitron-ai/alethia-mcp.

Policy commitments (true in v0.2; any future cloud features will be opt-in and clearly disclosed):

  • Zero telemetry collection by default. The runtime does not phone home, does not collect usage metrics, does not report crashes anywhere out of the box.
  • Opt-in cloud features. When the cloud dashboard / signed evidence service / agent observability layer ships, they will be explicit, separate, paid products you enroll in — not defaults that turn on silently.
  • Cryptographically signed evidence packs. Ed25519 keypair + canonical SHA-256 manifest. See Evidence. Signing happens locally; any public key registry is opt-in.

Evidence

The performance, safety, and patent claims are backed by reproducible evidence:

  • Benchmarks: click-assert-wait against Playwright, Puppeteer, Selenium, Cypress, TestCafe, Taiko. Latest results show Alethia at 12.9 ms avg, Playwright at 580.7 ms avg.
  • Validation suite: 12 jsdom-based test cases (6 runtime + 6 SDK contract) covering parser, policy, executor, kill switch, integrity hashing.
  • Signed evidence packs: Ed25519-signed canonical manifests via npm run evidence:keygen + npm run evidence:pack (in the closed alethia-core repo).
  • Patent filing: U.S. Patent Application No. 19/571,437 (non-provisional, in examiner queue), claiming priority to 63/785,814 filed April 9, 2025.

Design partners get access to the full alethia-core repo (including benchmarks, evidence pipeline, and patent docs).

Comparison to Playwright MCP

Microsoft ships Playwright MCP. Why use Alethia instead?

Playwright MCP Alethia
Architecture CDP marshalling, async-by-construction Zero-IPC, synchronous
Avg latency per step ~580 ms ~13 ms
Per-step safety gate none VITRON-EA1 fail-closed
Cryptographic audit none SHA-256 chained, Ed25519 signable
Telemetry Microsoft data flow by default off by default; opt-in only
Patent moat none US 19/571,437
Optimized for cross-browser public-web QA agent loops on localhost dev servers

Use Playwright MCP for cross-browser public-web testing where the human is debugging async test scripts.

Use Alethia for AI agents driving localhost dev servers in tight decide-act-verify loops where every millisecond and every fail-closed guardrail matters.

The two tools serve different jobs. Playwright will not catch up on the agent-loop wedge because closing the gap would require abandoning their cross-browser story (they'd have to become an in-process, Chromium-only runtime — i.e. become Alethia).

Patent Notice

Alethia practices a method that is the subject of:

  • U.S. Patent Application No. 19/571,437 (non-provisional)
  • Claiming priority to U.S. Provisional Application No. 63/785,814 (filed April 9, 2025)
  • Title: "Deterministic Local Automation Runtime with Zero-IPC Execution, Offline Operation, and Per-Step Policy Enforcement"
  • Status: Patent Pending — U.S. Patent and Trademark Office

The MIT license on the alethia-mcp bridge does not grant any patent license under U.S. Application No. 19/571,437 or any other vitron.ai patent rights. The closed Alethia Core runtime is governed by separate terms; commercial and production use may require a patent license once granted.

For licensing inquiries: gatekeeper@vitron.ai

Links

Alethia is built by vitron.ai. Patent Pending. Local-first. Built for the agents that build software.

About

No description, website, or topics provided.

Resources

Readme

License

MIT license
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

1 tags

Packages

 
 
 

Contributors 1