Engineer the execution layer for AI agents.
MeshHub turns APIs into agent-executable tools. Agents choose → MeshHub executes → results return. We're building the infrastructure between agent runtimes and real-world services.
Agents can reason. They can't execute reliably.
APIs are inconsistent (auth, schemas, errors, rate limits).
Tool execution needs standard schemas + predictable runtime.
Reliability matters more than demos (timeouts, retries, observability).
Agents need discovery + safe execution paths.
Transactions need user confirmation, not trust.
What MeshHub is
- An execution layer that makes APIs callable as agent tools.
- A shared registry + runtime for tool discovery and execution.
- A neutral layer: works with any agent runtime (OpenAI/Claude/Gemini/etc.).
What MeshHub is NOT
- Not an agent builder / agent framework.
- Not just an API directory.
- Not an enterprise API gateway like Kong/Apigee.
Kong manages API traffic. RapidAPI lists APIs. MeshHub executes tools for agents.
The engineering challenges you'll own
Schema generation
API → Tool schema generation (OpenAPI → JSON Schema, tool defs)
Auth adapters
API keys, OAuth, service accounts — secure secret handling
Execution runtime
Timeouts, retries, backoff, idempotency
Observability
Structured logs, traces, metrics per tool call
Quality scoring
Tool health checks, verification, reliability score
Developer experience
Sandboxed tool testing, playground, examples
The engineer we want
- You've built or maintained backend/platform infrastructure.
- You think in systems: reliability, failure modes, rate limits, security.
- You ship pragmatic solutions (not endless architecture debates).
- You can read messy APIs and make them usable.
- You care about developer experience and clarity.
Strong signals on your GitHub
- •Repos with gateways, proxies, routers, orchestrators
- •OpenAPI / schema tooling
- •Retries, timeouts, auth middleware
- •Observability / instrumentation
Not a fit if
- Mostly frontend/UI work
- Mostly prompt engineering / chatbot demos
- Prefer closed, slow enterprise processes
First 90 days
Harden tool execution runtime (timeouts, retries, error normalization).
Improve API → tool schema generation and validation.
Build verification: daily tool health checks + reliability score.
Improve DevEx: demo → playground → quickstart conversion loop.
Add guardrails for transaction flows with user confirmation.
Why this matters
Agents are becoming a new distribution channel for services. The winning layer won't be another UI — it will be execution infrastructure that agents trust.