# Synsmarts

> Agent-native observability platform. We run a hosting business on it.

Most hosts stop at "your server is healthy, the problem is in your code, good luck." We don't. Synsmarts traces every request end-to-end — browser, load balancer, application runtime, every database query and cache call, and back. Our agents read all of that alongside your code and tell you which line is causing the problem.

If your agent runs in Claude Code, Cursor, or any MCP-compatible client, it operates Synsmarts the same way it edits your code — deploys, rollbacks, traces, logs, and code search are first-class tool calls, not screenshots of a dashboard.

## architecture

- **Substrate** — Kubernetes on AWS. Customers pick their reliability tier, from single-AZ at 99% up to multi-region active-active at 99.999%. The control plane is multi-AZ at every tier, because no tenant should ever go down because of *our* issues.
- **Telemetry** — ClickHouse-backed unified logs, metrics, and traces. 100% trace capture, no sampling. We log at a level of detail that lets us reconstruct any request, any deploy, any incident after the fact. The telemetry is the product, not a feature of it.
- **Orchestration** — Temporal workflows for every tenant lifecycle operation. Repeatable processes, not heroics. Workflows are the only substrate where agents can safely operate against shared state without losing it on retry.
- **Code intelligence** — per-tenant code RAG/DAG (ChunkHound), physically isolated per tenant. Payment data, session tokens, and PII never cross tenant boundaries — not even into the index an agent reads from.
- **Agent surface** — two MCP servers exposing 35 tools. Every system we build asks: can an agent use this? If the answer is no, the design is wrong.

## we own the rails

Other hosts buy monitoring tools and bolt them on. We built the platform *that the monitoring is*. Our agents read logs, traces, performance metrics, and your application code at the same time and synthesize a diagnosis with a solution. No human team can do that across hundreds of customers at scale. No monitoring tool does it today.

Monitoring tools show dashboards. We tell you the answer.

This is the moat. You can't bolt it on after the fact. You have to build for it from the ground up.

## what we are not

- Not a monitoring tool you point at someone else's hosting.
- Not a hosting product with an AI chatbot stapled on the side.
- Not a coding agent with a hosting affiliate program.

We are the substrate. The agent is a first-class operator on top of it.

## for agents

- [llms.txt](/llms.txt) — site index
- [README.md](/README.md) — this page in raw markdown
- **MCP server** — SSE at `/api/v1/mcp` on the platform API endpoint. Tenant API key auth, server-side tenant_id injection (your agent never handles tenant_id). ~35 tools across deploy, diagnostics, self-service, and billing. Scopes: admin, deploy, read-only. Streaming progress on long-running ops. Structured errors with correlation IDs.
- **REST API** — `/api/v1/...`, same Django backend. CLI, REST, MCP are peer interfaces.
- **machines@synsmarts.ai** — coordination beyond tool calls

## for humans

humans@synsmarts.ai