NavFlow

A data plane for AI agents: the right context about your systems, in one read.

NavFlow connects to the systems you already run — logs, metrics, deploys, databases, alerts — and serves AI agents a live, correlated view of them over the Model Context Protocol (MCP). An agent asks one question about one entity — what happened to checkout? — and gets every event that concerns it, across all sources, merged into a single time-ordered timeline. The agent doesn’t call several systems and stitch the results together; NavFlow does the correlation.

It works in the other direction too: triggers watch the data as it arrives, and when a condition fires NavFlow pushes the correlated timeline to subscribed agents — the agent wakes up already holding the context.

NavFlow is self-hosted: it runs on your machine or your server, and your data stays with you. A managed, multi-tenant cloud version is coming soon — same feature set, run by the NavFlow team.

Architecture

Today NavFlow ships as a small self-hosted service — a daemon (navflowd) that serves the HTTP API and a built-in web console, plus navflow-mcp, a thin MCP server that proxies agent tool calls to the daemon. The pipeline has four stages.

1. Ingest — a connector turns an external source into events. Poll connectors (GitHub, Prometheus, Postgres, Docker logs) fetch on an interval; push connectors (Vercel, OTLP, webhooks) receive over HTTP. Each event keeps its original payload.
2. Store — events are written to an embedded database (DuckDB); navflowd is the only writer.
3. Serve — agents and the built-in web console read the stored events: a correlated query over a view, an entity facet, a source’s schema. Agents read over MCP; the console uses the same HTTP API underneath.
4. Watch — triggers evaluate conditions over the data as events arrive. When a trigger fires, NavFlow delivers a webhook to subscribed agents.

Data model

• A source is a configured connector instance.
• An event has a source, timestamp, text line, typed fields, named labels, and a lossless payload.
• A label is a named axis on an event (service, repo, env). The label marked primary is the key.
• An entity is a (label, value) pair.
• The catalog is the registry of sources, views, and triggers.
• A view correlates sources into one timeline for a key.
• A trigger fires on a condition over a view; agents subscribe to be pushed when it does.

Full definitions are in Concepts.

Documentation

• Quickstart — install, ingest a source, connect an agent.
• Concepts — the data model in full.
• Connecting agents (MCP) — transports, clients, and the tool set.
• Connectors — available connectors and their configuration.
• Deployment — local and self-hosted (Docker) deployment.
• CLI reference — navflow up, navflow mcp, and the entry points.

Need something NavFlow doesn’t do yet?

Custom connectors, missing features, or questions about the managed version — email ashish@navflow.ai. We’re building against real use cases and reply quickly.