Prompt Tokenizer & Pricing Estimator

# Hermes Dispatch Tool — Prompt Tokenizer & Pricing Estimator
# Source: https://hermesdispatch.dev/tools/prompt-tokenizer-pricing-estimator/
# Description: Estimate tokens in a prompt and compare LLM API pricing across providers.
# License: MIT (generated by hermesdispatch.dev)
#
# INSTALL:
#   1. Save this file as ~/.hermes/hermes-agent/tools/prompt_tokenizer_pricing_estimator.py
#   2. Restart Hermes or run /reset in a session
#   3. The tool auto-registers if Hermes uses auto-discovery of tools/*.py
#
# MANUAL REGISTRY (if auto-discovery is off):
#   from tools.prompt_tokenizer_pricing_estimator import register
#   register()

import json
import math

DATA = {"last_updated": "2026-07-12", "note": "Prices are per 1M tokens in USD. Token counts are estimates (~4 chars/token or 1.3x words, whichever is higher). Confirm current rates on provider sites before committing to a budget.", "providers": [{"name": "OpenAI GPT-4o", "model": "gpt-4o", "input_per_1m": 2.5, "output_per_1m": 10.0, "cached_input_per_1m": 1.25, "request_fee": 0.0, "context_window": 128000, "best_for": "General agent tasks, vision, tool use"}, {"name": "OpenAI GPT-4o-mini", "model": "gpt-4o-mini", "input_per_1m": 0.15, "output_per_1m": 0.6, "cached_input_per_1m": 0.075, "request_fee": 0.0, "context_window": 128000, "best_for": "Cheap high-volume agents"}, {"name": "Anthropic Claude 3.5 Sonnet", "model": "claude-3-5-sonnet-20240620", "input_per_1m": 3.0, "output_per_1m": 15.0, "cached_input_per_1m": 3.75, "request_fee": 0.0, "context_window": 200000, "best_for": "Long context, reasoning, coding"}, {"name": "Anthropic Claude 3 Haiku", "model": "claude-3-haiku-20240307", "input_per_1m": 0.25, "output_per_1m": 1.25, "cached_input_per_1m": 0.03, "request_fee": 0.0, "context_window": 200000, "best_for": "Fast, cheap agent calls"}, {"name": "Google Gemini 1.5 Pro", "model": "gemini-1.5-pro", "input_per_1m": 3.5, "output_per_1m": 10.5, "cached_input_per_1m": 0.0, "request_fee": 0.0, "context_window": 2000000, "best_for": "Massive context, multimodal"}, {"name": "Google Gemini 1.5 Flash", "model": "gemini-1.5-flash", "input_per_1m": 0.35, "output_per_1m": 1.05, "cached_input_per_1m": 0.0, "request_fee": 0.0, "context_window": 1000000, "best_for": "Low-latency, high-volume"}, {"name": "Groq Llama 3.1 70B", "model": "llama-3.1-70b-versatile", "input_per_1m": 0.59, "output_per_1m": 0.79, "cached_input_per_1m": 0.0, "request_fee": 0.0, "context_window": 131072, "best_for": "Fast open-weight inference"}, {"name": "Together AI Llama 3.1 70B", "model": "llama-3.1-70b", "input_per_1m": 0.9, "output_per_1m": 0.9, "cached_input_per_1m": 0.0, "request_fee": 0.0, "context_window": 131072, "best_for": "Open-weight fine-tuning hub"}, {"name": "Cohere Command R+", "model": "command-r-plus", "input_per_1m": 2.5, "output_per_1m": 10.0, "cached_input_per_1m": 0.0, "request_fee": 0.0, "context_window": 128000, "best_for": "RAG and enterprise search"}]}

def _ok(result):
    return json.dumps({"success": True, "data": result}, indent=2)

def _err(message):
    return json.dumps({"success": False, "error": message}, indent=2)


def estimate_tokens(text):
    if not text:
        return 0
    chars = len(text)
    words = len(text.strip().split())
    return math.ceil(max(chars / 4.0, words * 1.3))


def cost_for_provider(p, input_tokens, output_tokens, cache_hit=0.0, requests=1):
    cached = cache_hit * (p.get("cached_input_per_1m") or p["input_per_1m"])
    uncached = (1 - cache_hit) * p["input_per_1m"]
    in_cost = (input_tokens / 1_000_000) * (cached + uncached)
    out_cost = (output_tokens / 1_000_000) * p["output_per_1m"]
    per_request = in_cost + out_cost + (p.get("request_fee") or 0)
    return {
        "provider": p["name"],
        "model": p.get("model", ""),
        "per_request": round(per_request, 6),
        "monthly": round(per_request * requests, 2),
        "best_for": p.get("best_for", "")
    }


TOOL_NAME = "prompt_tokenizer_pricing_estimator"
TOOLSET = "agents"

SCHEMA = {
  "type": "function",
  "function": {
    "name": "prompt_tokenizer_pricing_estimator",
    "description": "Estimate tokens in a prompt and compare LLM API pricing across providers.",
    "parameters": {
      "type": "object",
      "properties": {
        "prompt": {
          "type": "string",
          "description": "The prompt text to tokenize and price."
        },
        "input_tokens": {
          "type": "integer",
          "description": "Optional override for input token count."
        },
        "output_tokens": {
          "type": "integer",
          "description": "Output tokens per request."
        },
        "requests_per_month": {
          "type": "integer",
          "description": "Number of requests per month."
        },
        "cache_hit_percent": {
          "type": "number",
          "description": "Percentage of input tokens that hit a context cache (0-100)."
        }
      },
      "required": ["prompt"]
    }
  }
}

def _run(args):
    prompt = args.get("prompt", "")
    input_tokens_override = args.get("input_tokens")
    input_tokens = int(input_tokens_override) if input_tokens_override is not None else estimate_tokens(prompt)
    output_tokens = int(args.get("output_tokens", 500))
    requests = int(args.get("requests_per_month", 100000))
    cache_hit = max(0.0, min(100.0, float(args.get("cache_hit_percent", 0)))) / 100.0

    results = [cost_for_provider(p, input_tokens, output_tokens, cache_hit, requests) for p in DATA["providers"]]
    results.sort(key=lambda x: x["monthly"])

    return _ok({
        "estimated_input_tokens": input_tokens,
        "output_tokens_per_request": output_tokens,
        "total_tokens_per_request": input_tokens + output_tokens,
        "requests_per_month": requests,
        "cache_hit_percent": cache_hit * 100,
        "cheapest_provider": results[0]["provider"] if results else None,
        "provider_costs": results
    })

def HANDLER(args):
    try:
        return _run(args)
    except Exception as e:
        return _err(str(e))


def register():
    """Manual registry hook. Import and call this to register with Hermes."""
    try:
        from tools.registry import registry
        registry.register(
            name=TOOL_NAME,
            toolset=TOOLSET,
            schema=SCHEMA,
            handler=HANDLER,
        )
    except ImportError:
        print("Hermes registry not found; skipping manual registration.")

if __name__ == "__main__":
    # CLI smoke test
    print(HANDLER({"prompt": "You are a helpful coding assistant. Explain the code the user pastes, then suggest three concrete improvements with brief rationale."}))