OpenAI demos Codex-powered vendor risk agent in ChatGPT
OpenAI's "workspace agents" feature for ChatGPT lets teams build Codex-powered agents that automate vendor due diligence — screening for sanctions, financial, and reputational risk and producing a structured report in minutes.
Score breakdown
Teams can automate structured, multi-step compliance workflows like vendor due diligence directly inside ChatGPT, with full run-trace visibility and no engineering overhead.
- 01OpenAI introduced workspace agents in ChatGPT — Codex-powered agents built for teams.
- 02The demo agent, called Trove, is a third-party risk manager modeled on an agent OpenAI's own finance team uses today.
- 03Trove screens vendors for sanctions, financial, and reputational risk, then produces a structured report.
OpenAI published a guided walkthrough of workspace agents in ChatGPT, a capability designed for teams that want to deploy Codex-powered agents without requiring dedicated engineering resources. The video, presented by Hojun, centers on building an agent called "Trove" — a third-party risk manager described as similar to an agent OpenAI's own finance team uses today to accelerate vendor due diligence with greater consistency and control.
The build process begins with a natural-language prompt describing the desired workflow, tools, skills, and systems the agent will need.
The build process begins with a natural-language prompt describing the desired workflow, tools, skills, and systems the agent will need. A "skill" — containing best practices, instructions, and metadata specific to the finance team's vendor risk assessments — is added to give the agent domain-specific guidance. ChatGPT then outlines and implements a plan for Trove, with the user iterating in a left-hand chat pane while the agent's tools, skills, and app configurations are assembled on the right. The result is a fine-tuned instruction set produced without manual engineering work.
Once built, Trove can be previewed in the same interface. The agent gathers evidence, runs a risk assessment using the embedded skill, orchestrates work across connected systems, and delivers a final structured report ready for human analyst review — all within minutes. Run traces showing tool calls, inputs, and decisions are visible throughout, giving the analyst full visibility into the agent's reasoning and actions.
Key facts
- 01OpenAI introduced workspace agents in ChatGPT — Codex-powered agents built for teams.
- 02The demo agent, called Trove, is a third-party risk manager modeled on an agent OpenAI's own finance team uses today.
- 03Trove screens vendors for sanctions, financial, and reputational risk, then produces a structured report.
- 04Agents are built via natural-language prompts; no technical engineering resources are required.
- 05A reusable 'skill' — containing best practices, instructions, and metadata — can be added to guide the agent's assessments.
- 06Users can monitor agent runs via full trace visibility, including tool calls, inputs, and decisions.
- 07The agent completes vendor due diligence in minutes, compared to a manual time-intensive workflow.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Apr 25, 2026 · 21:38 UTC. How this works →