# Volt HQ

The compute price oracle for AI agents.

## What it does

- **Compares pricing** across 9 providers (OpenAI, Anthropic, Groq, Together AI, DeepInfra, Fireworks AI, Cerebras, Hyperbolic, Akash) — 200+ offerings with live API pricing
- **Recommends optimal routing** — tells your agent where to get the same quality for less, with savings estimates
- **Tracks spend and budgets** — spending summaries by provider/model, savings reports, and threshold alerts

## Install

Auto-configure Cursor and Claude Desktop in one command:

```
npx volthq-mcp-server --setup
```

Detects installed clients, merges config without overwriting your existing MCP servers.

<details>
<summary>Manual setup</summary>

**Cursor** — add to `.cursor/mcp.json`:

```json
{
  "mcpServers": {
    "volthq": {
      "command": "npx",
      "args": ["-y", "volthq-mcp-server"]
    }
  }
}
```

**Claude Desktop** — add to `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "volthq": {
      "command": "npx",
      "args": ["-y", "volthq-mcp-server"]
    }
  }
}
```

</details>

## Tools

| Tool | Description |
|------|-------------|
| `volt_check_price` | Compare pricing across providers for a model |
| `volt_recommend_route` | Get optimal provider recommendation with savings estimate |
| `volt_get_spend` | Spending summary by provider and model (today/7d/30d) |
| `volt_get_savings` | Actual spend vs optimized spend comparison |
| `volt_set_budget_alert` | Set daily/weekly/monthly budget threshold alerts |

## Example

```
> volt_check_price { "model": "llama-70b" }

Price comparison for "llama-70b" — 14 offerings found
────────────────────────────────────────────────────────────
1. DeepInfra — Llama-70B
   Input: $0.20/M tokens | Output: $0.27/M tokens | Avg: $0.24/M
   Quality: 88% | Region: global

2. Hyperbolic — Llama-70B (FP8) on H100-SXM
   Input: $0.40/M tokens | Output: $0.40/M tokens | Avg: $0.40/M
   Quality: 85% | Region: global

3. Hyperbolic — Llama-70B (BF16) on H100-SXM
   Input: $0.55/M tokens | Output: $0.55/M tokens | Avg: $0.55/M
   Quality: 88% | Region: global

4. Groq — Llama-70B
   Input: $0.59/M tokens | Output: $0.79/M tokens | Avg: $0.69/M
   Quality: 88% | Region: global

5. Fireworks AI — Llama-70B
   Input: $0.90/M tokens | Output: $0.90/M tokens | Avg: $0.90/M
   Quality: 88% | Region: global

6. Together AI — Llama-70B
   Input: $0.88/M tokens | Output: $0.88/M tokens | Avg: $0.88/M
   Quality: 88% | Region: global

7. Akash — Llama-70B (FP8) on H100-SXM
   Input: $3.49/M tokens | Output: $8.72/M tokens | Avg: $6.11/M
   Quality: 85% | Region: global

8. Akash — Llama-70B (FP8) on A100-80GB
   Input: $5.24/M tokens | Output: $13.11/M tokens | Avg: $9.18/M
   Quality: 85% | Region: global

Cheapest is 97% less than most expensive option.
```

DeepInfra at $0.24/M, Hyperbolic at $0.40/M, Groq at $0.69/M, Fireworks AI at $0.90/M — all vs GPT-4o at $6.25/M.

## Supported providers

- **OpenAI** — GPT-4o, GPT-4o-mini
- **Anthropic** — Claude Sonnet 4.6, Claude Haiku 4.5
- **Groq** — Llama-70B, Llama-8B, Mixtral-8x7B
- **Together AI** — Llama-70B, Llama-8B, DeepSeek-V3
- **DeepInfra** — 75+ models with live API pricing (Llama, DeepSeek, Qwen, Mistral, Gemma, and more)
- **Fireworks AI** — Llama-70B, Llama-8B, DeepSeek-V3
- **Cerebras** — Llama-3.1-8B, Qwen-3-235B
- **Hyperbolic** — DeepSeek-V3, DeepSeek-R1, Llama-70B, Llama-8B
- **Akash** — Llama-70B, Llama-8B on H100 and A100 (live GPU pricing)

## Telemetry

Volt collects anonymous usage metadata by default to improve routing recommendations. This includes: provider name, model name, tool response time, and success/failure status.

**What is never collected:** prompts, outputs, API keys, token counts, or any user-identifiable content. IPs are hashed and truncated server-side.

To opt out, set the environment variable:

```
VOLT_OBSERVATIONS=false
```

## Links

- npm: [volthq-mcp-server](https://www.npmjs.com/package/volthq-mcp-server)
- Web: [volthq.dev](https://volthq.dev)

## License

MIT