AIG for agents & developers

AI infrastructure evidence your software can actually use.

Query the same model, hardware and compatibility evidence behind Agent Infra Guide’s public tools—without scraping articles or reverse-engineering prose.

Live remote MCP endpoint

https://mcp.agentinfraguide.com

Streamable HTTP · no API key required for the current public evidence tools

Connect AIG to your AI agent →Check a model on your hardware
AI infrastructure evidence process connecting dated primary specifications to transparent calculator assumptions and real-world workload validation
Evidence workflow. Documentation, calculated estimates and measured observations must remain distinguishable.

Seven tools, ready now

Give your agent infrastructure-specific capabilities

01get_model_requirements

Retrieve artifact size, context and source evidence for a model.

02find_models_for_hardware

Find models that fit a selected hardware memory envelope.

03check_model_compatibility

Return the planning result for a model and hardware combination.

04estimate_vram

Return the planning result for a model and hardware combination.

05compare_hardware

Return the planning result for a model and hardware combination.

06estimate_hosting_cost

Return the planning result for a model and hardware combination.

07compare_local_vs_cloud_cost

Return the planning result for a model and hardware combination.

Model requirements

/api/v1/models

Artifact size, context window, model family, source and evidence date.

Use this resource →

Hardware catalog

/api/v1/hardware

Unified memory and GPU VRAM records used by the compatibility engine.

Use this resource →

Compatibility matrix

/api/v1/compatibility

Precomputed model × hardware planning results with an explicit 25% reserve.

Use this resource →

Agent manifest

/agent-tools.json

Machine-readable discovery document for AIG's current agent-facing resources.

Use this resource →

New model evidence

Use the expanded compatibility dataset

The canonical model API now includes Qwen2.5-Coder 7B/14B/32B, Llama 3.2 1B/3B and Phi-4 14B. Use the API for software or open the evidence pages for the human-readable sizing context.

Qwen2.5-Coder hardware →Llama 3.2 hardware →Phi-4 hardware →Read model API →

Example workflow

From “Can it run?” to inspectable evidence

A client can read the compatibility matrix, match a model and hardware record, then show the planning result alongside the assumptions and underlying evidence.

GET /api/v1/compatibility

{
  "reservePct": 25,
  "methodology": "artifact size plus explicit reserve",
  "checks": [ ... ]
}

What you can build

One evidence layer, multiple products

✓

Answer model-fit questions inside an AI assistant

✓

Add evidence-backed compatibility data to a hardware guide

✓

Build model selectors and infrastructure recommendation tools

✓

Use one canonical dataset across scripts, agents and websites

Evidence, not a black box

Built to expose the assumptions.

Compatibility is currently a planning estimate based on model artifact size plus an explicit reserve. It is not presented as a runtime benchmark. Source and checked-date fields stay with the underlying records so clients can judge freshness and provenance.

Current interface

Static JSON endpoints · free to read · versioned under /api/v1

The live remote MCP server and JSON endpoints use this same canonical evidence layer.

Need the human-facing version?

Use the interactive checker to select a model and hardware configuration, or embed it on another site with attribution.

Check if your hardware can run your model →Compare local vs cloud cost →