Use EdgeAIStack inside your engineering workflow
Every EdgeAIStack sizing engine is callable programmatically — an OpenAPI 3.1 spec for GPTs, scripts and custom agents, and an MCP server for Claude, Cursor and Windsurf. Same vendor-neutral engines that power the on-site calculators, returning structured JSON you can drop into a deployment pipeline.
Base URL https://api.edgeaistack.ai. All tools are
POST endpoints that take a JSON body and return a structured result.
The machine-readable spec is public (no auth required to read):
Available tools
| Tool | Endpoint | Returns |
|---|---|---|
| Recommend platform | POST /api/v1/tools/recommend_platform | Ranked best-fit platforms with alternatives |
| Design system | POST /api/v1/tools/design_system | Full deployment design, sized infrastructure and BOM |
| Calculate streams | POST /api/v1/tools/calculate_streams | Concurrent stream capacity for a platform/workload |
| Camera stream capacity | POST /api/v1/tools/camera_stream_capacity | Full pipeline as constraints (decode, encode, inference, memory, network) with NVIDIA datasheet source lines, bottleneck and headroom |
| Model memory fit | POST /api/v1/tools/model_memory_fit | Does this LLM/VLM/ASR model fit in this module's memory: weights, KV cache, runtime overhead and OS/reserve breakdown, verdict, headroom ladder and sources |
| Jetson configuration checker | POST /api/v1/tools/jetson_configuration_checker | Module × JetPack/L4T × install method × runtime × power mode: per-check status, compute stack, ordered verify commands with expected output, remediations, known issues and sources |
| Benchmark explorer | POST /api/v1/tools/benchmark_explorer | Matching benchmark rows for a hardware × model × precision × runtime query, each with evidence class and source, plus the expected range per precision, a cross-hardware comparison and the class-D fallback chain when nothing was measured |
| Benchmark reality check | POST /api/v1/tools/benchmark_reality_check | Measured fps/latency/tokens-per-s vs. the Benchmark Explorer expected range: verdict (within/below/above range or no reference), gap ratio, causes ranked by fit to the gap with verify commands and remediation, and the Jetson Configuration Checker hand-off when JetPack and power mode are given |
| Estimate bandwidth | POST /api/v1/tools/estimate_bandwidth | Network bandwidth for a camera set |
| Estimate power | POST /api/v1/tools/estimate_power | Power draw and PoE/supply budget |
| Estimate storage | POST /api/v1/tools/estimate_storage | Storage capacity and drive endurance |
Have a real benchmark measurement? Submit it to
POST /api/v1/measurements/submit (no API key required) with your
hardware, model, precision, runtime, the exact command you ran and your attribution — see the
measurement protocol for the field list and
an example body. A human reviews each submission against the protocol before it becomes a
class-C row in the benchmark database.
Tool endpoints require an API key sent as a header:
X-API-Key: <your-key>. Keys are issued on request —
email [email protected]
with a line about your use case. The openapi.json document itself
is open for discovery.
Rate limits
Per client IP, per hour: agent tool endpoints
(/api/v1/tools/*) 50 requests; the key-free calculator endpoints
(/api/v1/{tool-name}, same engines, public envelope) 100 requests;
the assistant 30 requests (plus a 5/minute burst cap). Exceeding a limit returns
429 with a retry hint. Need more for a production integration?
Say so in your key request.
The MCP server exposes the same six tools to MCP-capable clients (Claude Desktop, Cursor, Windsurf). It runs locally over stdio and calls the hosted API. Add it to your client's MCP configuration:
Point EDGEAISTACK_API_URL at
https://api.edgeaistack.ai. Once connected, your assistant can recommend
platforms, size power/storage/bandwidth and design full deployments without leaving the chat.
Every tool above has an interactive calculator with the assumptions shown: Hardware Selector, System Designer, Power Budget, Storage Endurance, Network Bandwidth, and the full engine list. See the methodology for how the numbers are derived.