Compare three proposals. Keep every conclusion tied to its source.
Extract the price, delivery time, revisions, and support. Compare by the priority you choose.
Keep this tab open. Only approved tasks will arrive here.
Open the invitation on your other device, signed into this private site with the same account. Approve its fingerprint here before sharing tasks.
Uses PeerJS for connection metadata and WebRTC for task traffic. Connectivity depends on your network.
This invitation grants no task access until the host approves. Do not approve devices you do not trust with the source text.
Paste a proposal using labeled fields. Missing fields stay unknown. This prototype reads USD prices and delivery in days.
Working: source-linked field extraction, worker tabs, cross-device WebRTC transport, host approval, signed primary/mirror grants, task-bound intent hashes, rejection after revocation, timeout recovery, and signed evidence export.
Optional: Qwen 0.5B through WebLLM for a local draft. It downloads a complete small model onto this device. Hardware compatibility and download availability vary.
Next stage: SwarmLLM layer splitting, production AXIOM protocol integration, durable authority service, model-compute verification, and GPU-aware scheduling. No distributed-model speedup is claimed here.
The host is the authority and a single point of failure. Revocation blocks future dispatch and result acceptance at the host; it cannot erase data already received. An approved peer can still return incorrect computations. The host checks deterministic extraction against the source in this prototype.
Evidence verifies integrity against its signed checkpoint. Preserve a trusted checkpoint separately to detect replacement or rollback of an entire export. This browser protocol uses ECDSA P-256 and is not wire-compatible with AXIOM’s Python HMAC implementation.
AXIOM bonded-pair source ↗ · SwarmLLM ↗