Skip to main content
BEN STONE

Trusted Execution

Conduit → Halo

Audited headless browser + headed policy/approval runtime

If an agent can browse and act, the session needs a proof trail stronger than a screenshot — and headed runs need policy and human approval, not stealth myths.

Visit Site (opens in a new tab)
Status
Reference Implementation
Role
Builder / product author
System Type
Auditable browser lineage (headless + headed)

Capabilities

  • Conduit — headless browser with SHA-256 audit chain + Ed25519 proofs
  • Exportable / verifiable session proofs
  • MCP server / PyPI conduit-browser
  • Conduit Halo — headed Chromium with policy engine and human approvals
  • Action/screenshot logs with redaction; Observe→Act→Verify tab orchestration
  • Explicitly not stealth / CAPTCHA-bypass / payment auto-submit

Problem

Agent browsers can erase the audit trail. Headed automation without policy turns “watching the agent” into uncontrolled action. Both need trustworthy execution — different constraints.

What I Built

Conduit is a headless browser with a cryptographic audit layer: actions form a hash chain and can be signed so sessions produce exportable proof.

Conduit Halo (from conduit-halo README only) is a standalone visible AI browser runtime: Playwright headed Chromium, policy engine (allow/block, approval triggers, session limits), human approvals, and redacted action/screenshot logs. No direct Conduit imports — adapter interfaces only. It is not a “score ring” brand myth.

ConduitScore — AI visibility SaaS — is a separate live product and has its own card.

How It Works

Headless path: agent drives Conduit → chained audit → signed export. Headed path: Halo runs visible sessions under policy with approval gates and recorded evidence. SwarmSync uses Conduit as a verification oracle for delivery checks.

SOURCE
Agent intent
Conduit (headless)
Headless browser
Action log
SHA-256 hash chain
Ed25519 signed proofs
Exportable session proof
Conduit Halo (headed)
Playwright headed Chromium
Policy engine
Human approvals
Redacted action / screenshot logs
Observe → Act → Verify

Why It Matters

Trusted agent execution needs integrity of what happened — not narration. Separating ConduitScore keeps visibility SaaS from being confused with the browser runtime.

Technical Depth

Conduit: reference implementation + active development, production-adjacent in SwarmSync/Genesis. Halo: active development / working prototype monorepo (apps + demos). Halo explicitly excludes stealth, CAPTCHA bypass, proxy rotation, and payment auto-submit.

Status

Reference implementation (Conduit) with active development on both runtimes. Visit Site points to the SwarmSync Conduit product page — Halo has no public marketing site.

Evidence

Conduit product page and PyPI package are linked. Halo descriptions are limited to conduit-halo README/docs facts.