ProxVantaDiscover, tune, and run agents with shared context.
ProxVanta gives teams a home for public agent discovery, private forks, organization-wide collaboration, and portable markdown specs you can run locally with use pv or carry cleanly into ChatGPT/Codex, Claude, Figma, and other MCP-capable clients by pasting the agent markdown into a file and asking your prompt to reference it.
Discover public agents
Browse specialists published by other teams, from design reviewers to product owners and sanitizers.
Fork and tune privately
Install a public agent into your organization, rename it, change the markdown brief, and keep your private version aligned to your team.
Run them anywhere
Run the same agent locally with `use pv`, or copy the generated markdown into ChatGPT/Codex, Claude, Figma, or another MCP-capable client and reference it in your prompt.
How it works
Built for the real way teams adopt agents.
ProxVanta is not another isolated chat surface. It gives teams a shared structure for agent discovery, private customization, public publishing, and portable execution across clients. Run agents locally with `use pv`, or copy the markdown into a file and tell your prompt to use it.
Shared source of truth
One agent definition. Three execution paths.
Write the brief once, then use `use pv` locally or export client-aware variants for ChatGPT/Codex, Claude, and a generic markdown path. Your team does not have to maintain three disconnected prompts.
- Shared markdown source with client-specific preview and copy output
- Tools, strengths, and response structure kept in sync with the portable spec
- Public and private agent pages that teams can actually share
Team-ready workflows
Structure around organizations, not isolated chats.
ProxVanta gives teams a place to discover agents, publish improvements, invite collaborators, and keep usage tied back to a shared organization context.
- Public organization pages with published agents
- Invite-first onboarding with team management and pending invite tracking
- Recently used agents and team-level usage visibility
Agent context
Code, backlog, and design stay connected.
ProxVanta helps teams define product and repository context so agents can operate with less ambiguity and better handoffs across implementation, backlog, and design.
- Repository and product structure that maps to how teams actually work
- Figma, GitHub, and Linear-aware agent specs without locking runtime to the server
- Clear, portable context for local-first agent execution
Business-ready defaults
Twenty built-in agents for actually running a business.
ProxVanta now starts with a full business-first catalog. Use use pv locally for a purpose-built teammate, or copy the markdown into a file and tell your prompt to reference it in ChatGPT/Codex, Claude, Figma, or any MCP-capable client.
- Engineer
- Designer
- Product Owner
- QA
- Security
- Operations
- Customer Support
- Executive Assistant
- Recruiting
- Compliance
- Growth
- Sales
- Marketing
- Content
- Partnerships
- Finance
- Analytics
- Research
- Legal
- Sanitizer
Local-first examples
Real `use pv` prompts your team can run locally.
The command shape stays simple. Pick the agent, add the business problem, and let the collaborator brief do the rest.
Operations
use pv operations redesign this workflow so the handoffs, owners, and follow-ups are explicitSales
use pv sales review this offer and tell me where prospects will hesitate or push backFinance
use pv finance evaluate the budget, margin, and downside exposure in this pricing changeSecurity
use pv security review this auth flow and flag the highest-risk trust boundary problemsExecutive Assistant
use pv executive-assistant turn this messy set of priorities into a realistic plan for the weekCustomer Support
use pv customer-support rewrite this support response so it is clearer, calmer, and more usefulStart with the shared spec
Bring your team into one cleaner agent workflow.
Discover public agents, install private forks, tune them for your organization, and carry the same spec into `use pv`, ChatGPT/Codex, Claude, Figma, and other MCP-capable tools by running it locally or referencing the markdown file directly.