LegitMate
One click deploy specialized agents pre-configured with relevant skills, tools and plugins on cloud.
Demo
About
LegitMate turns a specialist brief into a configured, owned Hermes workspace. A user chooses one of 279 Agency specialists, reviews its identity files, selects an approved model and capabilities, deploys the workspace, and verifies that the intended configuration reached the running instance.
How it's made
The frontend uses Next.js, TypeScript, Shadcn, and Base UI. It gives people one clear five-step flow: search 279 Agency specialists, inspect the chosen profile, pick a model, add approved capabilities, and review the exact configuration before deploying. Agent37 handles the infrastructure. LegitMate provisions an isolated Hermes workspace, writes the selected SOUL.md, USER.md, AGENTS.md, default skills, plug-ins and runtime configuration to fixed server-owned paths. Users can reopen their owned instances and access chat, terminal, files, integrations, settings, lifecycle controls, and the Hermes dashboard. LLM can be API or BYOK (Bring your own Key and base url) Privy provides authentication and the embedded wallet identity used to scope every instance and action to its owner. The browser never supplies a trusted user ID, and secrets stay on the server. Funding and withdrawals are deliberately disabled unless the chain, asset, recipient, and production policy are explicitly configured. The Graph gives deployed specialists live Uniswap V3 market data with block-level source details, TVL, volume, and pool evidence. The agent’s analysis is shown beside that evidence instead of presenting an uncited AI answer. Surplus model discovery also runs server-side, so available models can be added without exposing provider credentials; the default Hermes models remain available if Surplus is not configured. The hackiest useful part is launch reconciliation. Agent37 can finish creating a workspace after a reverse proxy times out, so I attach a client request ID, search only the authenticated owner’s instances, and resume verification instead of blindly creating a duplicate. I also added a server-owned, tool-disabled readiness task so the product can prove the deployed specialist actually responds without accepting arbitrary prompts or commands. Everything runs on a VPS behind Nginx and Cloudflare, with GitHub Actions deploying pushes to main. Vitest, Playwright, TypeScript, ESLint, production builds, configuration receipts, and live health checks back the release. The result is not another agent picker or a mock dashboard. It is a practical way to choose specialist expertise, deploy it into infrastructure you own, and verify what actually landed.