How it works

Three steps.
One analyst.

ServiceHop sits alongside your existing stack. It reads traffic, models it, and puts an AI analyst — the Analyst — on top of the model who can investigate your system on demand. No agents in the critical path. Nothing in your services. Your resident analyst.

01
Ingest

Point your traffic at ServiceHop.

ServiceHop receives API traffic through a thin forwarder or a proxy tap — whichever is less intrusive to your stack. Minutes to set up using ServiceHop's CLI or API. No SDK in your services. Nothing in your critical path.

02
Model

A living structural model of your system.

Traffic is profiled structurally and semantically. Fields are classified by purpose, endpoints are fingerprinted, services are discovered from the traffic itself — even the ones you forgot were running — sessions are grouped, and flows are reconstructed end-to-end across services without distributed tracing. The result is a model of what your system actually is, kept current from real calls.

03
Investigate

Ask the Analyst.

You ask a question about your architecture. The Analyst plans an investigation, runs real tools against the model and previous findings, and returns new findings with the calls that justified them. Cacheability opportunities, traffic concentration, structural waste, drift between snapshots — each traceable, each auditable, each drillable back to the raw traffic.

What it sees

Four primitives.

ServiceHop reduces the noise of telemetry to four things that actually describe a system.

Calls

Every API call: headers, payloads, latency, response shape — sampled and classified.

Services

What each endpoint actually does, derived from how its traffic behaves.

Sessions

Related calls grouped into a single user or system journey.

Flows

Multi-service transactions reconstructed end-to-end from real traffic.

How to ask

Three modes of analysis.

The same model, asked three different questions.

01 · The Analyst
Ask a question. Get an investigation.

The Analyst plans the steps, runs real tools against the model, and returns a finding with the evidence it used. Every step is shown. Every number is traceable. You can drill into any piece of evidence and see the calls behind it. This is the mode that makes ServiceHop feel like an analyst, not a dashboard.

  • "Where is our complexity coming from?"
  • "What would caching unlock on /catalog?"
  • "What changed after the March release?"

Watch a real investigation →

02
Snapshot

A complete picture of the current living model — what exists, what matters, where the traffic concentrates, where the waste lives. The foundation the Analyst investigates on top of.

03
Comparison

Compare two snapshots. New flows, structural drift, shifts in traffic distribution, endpoints that started behaving differently. Useful before and after migrations, releases, or acquisitions.

Technical foundations

Nine things that make it work.

Not a list of buzzwords — the actual capabilities ServiceHop is built around.

Semantic classification

Request and response fields are typed by purpose — identifiers, queries, metadata, errors — not by name. Two services with different schemas can still be recognized as equivalent.

Flow reconstruction

Multi-service transactions rebuilt end-to-end from raw traffic. Topology, latency accumulation, and structural dependencies — without distributed tracing.

Service discovery

ServiceHop reconstructs the real service topology from live traffic — including services nobody remembers deploying. The model stays current as services appear, disappear, or shift traffic between versions. No CMDB required; the traffic is the source of truth.

Cacheability simulation

Replays real traffic against cache strategies before you ship them. Quantifies hit rate, response similarity, and the risk of incorrect reuse.

Evidence-backed findings

Every finding ships with the calls that justified it. Auditable, drillable, replayable — never a black-box recommendation.

Drift detection

Compare any two snapshots of the living model. New flows, vanished endpoints, shifts in traffic concentration, endpoints that started behaving differently — surfaced as a structured diff, not a wall of deltas. Before and after migrations, releases, or acquisitions become observable at the architecture level.

Deterministic

Facts and findings are computed from clearly-stated thresholds. AI is used for narrative summary only — never to invent the underlying numbers.

Self-hosted or managed cloud

Runs in your own infrastructure when compliance requires it — air-gap-compatible, nothing leaves your network. Or runs on a privacy-aware European hypervisor we operate for you, so you can be live in an afternoon. Same product, same data boundaries, your choice of where it lives.

Agentic investigation

The Analyst is an AI agent that plans investigations and runs real tools against the ServiceHop model. It is MCP-native — it speaks the same protocol as the agent stack you already use. The primitives are deterministic; the agent is the interface. Every investigation is traceable, auditable, and reproducible.

See a real
investigation.

Sample analysis · one question, start to finish

Watch an investigation →