Your AI Agent Can Now Power and Run VTX Bots
Your chosen agent can power Main, Review, and Screener, operate in either response mode, and work with both Server Mode and Client Mode. VTX continues to manage the bot lifecycle, prompts, safeguards, execution, billing, and tracking.
Choose how your agent contributes
Two response modes are available.
Provider response
Your agent behaves like a conventional AI provider. VTX sends the prompt and context, receives the model response, and processes it through the normal parser.
This is the familiar VTX workflow with your preferred agent supplying the intelligence.
Decision candidate
Your agent produces the structured trading candidate directly. VTX validates it against the same bot settings, risk controls, account constraints, and execution rules used for every other decision.
The agent gets more direct responsibility for the proposed decision without bypassing VTX safeguards.
Every decision still belongs to a normal VTX bot
Connecting an agent does not create a separate or invisible trading system.
A normal VTX Trader must be running before the agent can produce live bot decisions. Stopping the Trader stops new agent-powered decisions.
Both response modes work in either execution mode:
- In Server Mode, VTX runs the bot infrastructure and charges the usual completed-call fee.
- In Client Mode, there is no VTX platform inference fee, but an active browser, desktop app, or supported headless runtime must continue owning the bot.
The connected agent host must also remain active. If it disconnects or cannot complete a request, VTX fails safely instead of silently switching to another model.
Complete VTX tracking remains intact
Agent-powered decisions appear through the same VTX experience as decisions from other models.
Main and Review activity remains visible in the Trade-page AI panel and normal history. Screener evaluations remain available through the existing Screener experience.
VTX preserves the complete decision trail, including:
- The exact system and user prompts delivered to the agent
- The market, account, and strategy context
- The required output schema
- The response or decision candidate returned
- The parsed and validated result
- The selected model, agent, and reasoning effort
- Response time and token reporting
- Server fees and accepted outcomes
This gives users a clear record of what the bot saw, what the agent returned, and what VTX ultimately accepted. It also preserves comparable evidence for future research and trade-chain analysis.
Use the agent you already work with
The new agent-driven workflow is not limited to Codex.
Compatible harnesses can claim a VTX inference request, reason over the exact prompt and context, and return the completed result through the public VTX CLI. Each harness keeps its own first-party login or subscription session.
Codex users also retain the automated packaged host, which can run the Codex connection without manually operating the agent-driven loop.
Go from investigation to action
The same agent can use VTX Insights to understand how your bots are performing before you change anything.
It can reconstruct complete trade chains using effective settings, prompts, decisions, market context, orders, fills, fees, funding, and realized results. With separately approved permissions, it can then propose settings changes, configure the selected inference mode, start the normal Trader, and verify whether the bot is genuinely running.
Analysis, settings, bot control, profile management, connections, and trading remain separately authorized. Your agent receives only the access you approve.
Get started from one instruction
The public VTX Insights page contains the current setup, permissions, operating instructions, and live capability catalog.
Give your preferred agent this request:
Read the VTX Insights guide and help me connect this agent as a VTX inference host. Explain Provider response and Decision candidate, show me the proposed settings, and configure my bot in Server Mode or Client Mode. Start the normal Trader only after I approve, then verify that it is genuinely running.
From there, your agent can guide the connection, request only the required permissions, configure Main, Review, or Screener, and keep the complete workflow inside VTX’s normal safety and tracking boundaries.