Technical Preview

Route with confidence.
Never gamble on a charger.

EvoSoft Routing is a vendor-agnostic EV route-planning MCP surface that treats charging reliability — not just distance or time — as the thing worth optimising for. Deterministic routing and energy math, evidence- backed site scoring, and full provenance on every recommendation, with an LLM only ever used to interpret intent and explain trade-offs — never to invent a coordinate, a distance, or a charger's status.

Request Access See How It Works
LLM-invented distances or SOC numbers
0

Distances, SOC%, or charger facts ever invented by a model

Vendor lock-in
None

Tie to a specific map, charging, or model vendor

Route scoring approach
Deterministic

Route and charging-site scoring — transparent, not a model guess

Provenance
Full

Provenance on every significant recommendation

Why it's built this way

An LLM interprets intent. It doesn't calculate a route.

A model is genuinely good at turning "I don't want to trust isolated chargers, I'd rather lose 20 minutes than get stranded" into a structured objective. Coordinates, distances, SOC, and connector compatibility are a different kind of problem — deterministic code you can test and audit, never a judgment call left to the model.

  • Vendor-agnostic by design

    No map provider, charging network, vehicle brand, or LLM vendor is baked into the contract — every provider maps into one canonical route and charging-site model behind a stable interface.

  • Reliability over max kW

    A site is scored on operational confidence, stall redundancy, operator/backend diversity, and fallback coverage — twelve stalls sharing one power cabinet are not twelve independent options.

  • Resilience, not just feasibility

    A recommended route exposes fallback distance, single-point-of- failure count, and operator diversity — a route that's merely "feasible" isn't the same as one that survives one bad day.

  • Full provenance, from day one

    Every assertion — fact, inference, estimate, assumption, or LLM interpretation — is labelled and traceable back to a source, never blended into one undifferentiated "confidence" number.

  • Small, composable tools

    routing.plan for a simple client, or routing.find_charging_sites / routing.score_charging_site / routing.estimate_energy / routing.validate individually for a sophisticated agent that wants to orchestrate the pieces itself.

  • Honest about what's real today

    This is a technical preview: charging-site reliability scoring is live, backed by real Open Charge Map data. Route planning, energy estimation, and route validation are still honest stubs — every tool says so plainly in its response, never a silent guess.

How it works

Deterministic core, model-assisted edges

The planning pipeline resolves locations and a vehicle profile, retrieves candidate routes, discovers and scores charging sites, simulates energy usage leg by leg, then only reaches for a model to parse ambiguous intent or explain the result in plain language.

  1. Describe the mission

    Origin, destination, vehicle, and — optionally — plain language like "I'd rather lose 20 minutes than get stranded," which a model turns into structured objectives and a risk profile.

  2. Deterministic candidates, evidence-scored sites

    Candidate routes, charging sites along the corridor, and energy simulation all run as plain code against provider data — never a model guessing a distance or a charger's live status.

  3. A resilience-scored recommendation

    The result carries a route, alternatives, charging stops, confidence per dimension, and full evidence — a model only narrates why one plan was chosen over another.

Good to know

Common questions

Is this live yet?

The MCP contract and authentication are live today. Charging-site reliability scoring is real, backed by live Open Charge Map data — it's the part of the spec we chose to build first, since it's the actual differentiator. Route planning and energy estimation are still honest stubs, returning clearly-labelled placeholder data until a route provider lands.

Will the tool contract change once real providers land?

The public shape — routing.plan, routing.find_charging_sites, routing.score_charging_site, routing.estimate_energy, routing.validate, and the REST equivalents — is designed to stay stable; providers plug in behind it, they don't reshape it.

Does the LLM ever decide what's feasible?

No. Coordinates, distances, route geometry, charger availability, connector compatibility, energy feasibility, and scoring are always code, never a model's judgment call.

Who is this for?

Evo Copilot itself, other LLM agents and MCP clients, internal .NET services, REST/API clients, and — longer term — a higher-level Geo MCP that delegates road-feasibility and charging questions here.

Ready to plan
without gambling on a charger?

EvoSoft Routing is available by request. Tell us about your fleet or use case and we'll get you set up.

Request access