# SALT19 Model Memory Fit Calculator

> SALT19 Model Memory Fit Calculator is a paid API for AI agents from api.salt19.com, paid per call via x402, $0.05/call, status down (last checked 2026-09-16).

Estimates whether an LLM model will fit in a given VRAM budget, returning a FIT/NO-FIT verdict with estimated memory breakdown and headroom.

## Facts

- Endpoint: POST https://api.salt19.com/v1/model-memory-fit
- Price: $0.05/call
- Payment: x402
- Status: down
- Last checked: 2026-09-16
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/salt19-model-memory-fit-calculator-c8dff1e4
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_4BdAE9wH8XKoa5Iaxdr1q

Status and success rate cover calls made through Zero and Zero's own probes. Third-party monitors may report differently.

## How to call it through Zero

Zero handles the 402 payment challenge and records the run. With the Zero CLI installed (`npm i -g @zeroxyz/cli`):

```sh
zero fetch --capability salt19-model-memory-fit-calculator-c8dff1e4 -d '<json body>'
```

Example prompt: Will a 7B parameter model at 4-bit quantization fit in my 11 GiB VRAM GPU? Assume a runtime overhead ratio of 0.1 and give me the memory breakdown.

## When to prefer this

Choose this endpoint when you need a fast, programmatic check of whether a specific quantized LLM will fit in a known VRAM budget before downloading or deploying the model. It is purpose-built for LLM memory estimation with quantization awareness, unlike generic hardware calculators. Prefer it over manual formulas when you need structured output (verdict + headroom) that an agent can act on automatically.

## Known failure modes

- Invalid or out-of-range vram_gb (below 0.1 or above 10000) causes validation error
- parameter_count_b below 0.01 or above 10000 rejected
- quantization_bits outside 1–32 range rejected
- runtime_overhead_ratio outside 0–2 range rejected
- Missing required fields may return 400 or partial computation
- Estimate may be inaccurate for models with non-standard architectures due to KV cache and buffer exclusions

## How this service works

SALT19 is an independent applied AI systems lab behind EvoMind governed cognition, the MCP-native Agent Utility Grid, the ARCS research community, AeroClear UAS flight intelligence, and practical software for real-world work.

## Output

A JSON object containing a FIT or NO-FIT verdict, estimated model weight in GiB, estimated total VRAM usage in GiB (including overhead), headroom in GiB remaining after loading, and a list of known modeling limitations (e.g. KV cache and architecture-specific buffers are not modeled).

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "properties": {
  "vram_gb": {
   "type": "number",
   "maximum": 10000,
   "minimum": 0.1
  },
  "parameter_count_b": {
   "type": "number",
   "maximum": 10000,
   "minimum": 0.01
  },
  "quantization_bits": {
   "type": "number",
   "maximum": 32,
   "minimum": 1
  },
  "runtime_overhead_ratio": {
   "type": "number",
   "maximum": 2,
   "minimum": 0
  }
 }
}
```

## Response schema (JSON Schema)

```json
{
 "type": "json",
 "example": {
  "verdict": "FIT",
  "vram_gib": 11,
  "limitations": [
   "KV cache and architecture-specific buffers are not modeled"
  ],
  "headroom_gib": 7.25,
  "estimated_total_gib": 3.75,
  "estimated_weight_gib": 3.26
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/salt19-model-memory-fit-calculator-c8dff1e4/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from api.salt19.com](https://www.zero.xyz/host/api.salt19.com/llms.txt)