Documentation — SYNCRONIX

SYNCRONIX/Docs

Introduction

SYNCRONIX is an autonomous AI execution network designed for Kinetiq Elysium. You describe an outcome; an Orchestrator and four specialist agents plan it, verify it and execute it, then settle every approved action on-chain.

Public preview. The console runs a live AI research agent: real web search, real page reading, streamed progress and saved run history. It has no wallet and cannot send transactions. On-chain settlement is planned.

1.Intent — describe the outcome
2.Plan — Orchestrator builds the execution graph
3.Execute — agents work in parallel lanes
4.Verify — Sentinel simulates every action
5.Settle — Executor finalizes on Kinetiq Elysium

Quickstart

Give the live agent a research task from the console. Each run uses a real AI model and live web research, and is saved to run history.

1.Open the console on the Platform page.
2.Describe a research task, or keep the suggested one. Say what to find out and what a good answer looks like.
3.Press “Run agent” and follow the plan, the searches, the pages it reads and the report as it is written.
4.Open History to revisit any earlier run, including its steps and sources.

Open the console →

Intents

An intent is the outcome you want, written in plain language. Good intents name the assets involved, the limits that must not be crossed and what “done” means.

agentA specialist with a role, a set of tools and a scoped context.

intentThe outcome you want, with assets, limits and a definition of done.

graphThe Orchestrator's plan: tasks ordered into stages by dependency.

policySigned rules agents cannot exceed: spend, slippage, allowlists.

receiptThe final record of a run: every transaction, hash and cost.

Execution graph

Design: the Orchestrator turns an intent into a directed graph of tasks. Tasks without dependencies run in parallel; dependent tasks wait for the outputs they need. Failed checks send a task back with exact feedback and the graph is re-planned.

Agent runtime

Each agent runs as an isolated step: it receives its role, its task and only the context it needs. Every tool call is logged with inputs, outputs and timing so a run can be replayed and audited.

Settlement

In the planned live mode, approved transaction bundles are signed by Executor under your policy, submitted to Kinetiq Elysium and tracked to finality. The run ends with one receipt containing every hash, fee and outcome. This step is not available yet: the current agent researches and reports only.

Policies

Planned: policies will be enforced by Sentinel before anything is signed. If a simulated transaction breaks a rule it will be rejected and returned for revision. The example below illustrates the kinds of limits a policy expresses; it is not a published format.

json · illustrative

{
  "maxSlippageBps": 30,
  "maxSpendPerRun": { "asset": "USDC", "amount": "25000" },
  "allowlist": ["USDC", "ETH"],
  "requireSimulation": true
}

Status & roadmap

Available now

› Live AI research agent with web search and fetch
› Streamed progress for every step
› Run history and reports stored in PostgreSQL

Planned

› Live settlement on Kinetiq Elysium
› Wallet connection and signing policies
› Scheduled and triggered runs
› Programmatic access for developers

No public API or SDK has been released. Anything not listed under “Available now” should be treated as planned.

Glossary

Orchestrator, Scout, Architect, Sentinel and Executor are the five roles in the SYNCRONIX design; in the current preview a single research agent performs the Orchestrator and Scout roles. A lane is one agent's track of work inside a stage. A receipt is the auditable result of an intent.