io.github.helphub369/model-ruler

What it mainly works on, as our AI model read its tool descriptions: nothing outside itself: it only answers questions.

None of its 12 tools are described as changing anything.

• token-counter

  “Use when a user asks how many tokens a given text will consume, or needs to estimate prompt size before pricing a workload.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  3 inputs, 1 required

  • textstringrequired“Text content to count tokens for”
  • tokenizerstring“Tokenizer family (default: default)”
  • expected_out_tokensnumber“Expected output token budget (optional)”
• provider-cost-calculator

  “Use when a user asks what an LLM workload costs on a specific provider/model, or wants to compare cost across providers.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  6 inputs, 2 required

  • tokens_innumberrequired“Input tokens per call”
  • tokens_outnumberrequired“Output tokens per call”
  • calls_per_monthnumber“Monthly call volume”
  • providerstring“Target provider (e.g.”
  • modelstring“Target model (e.g.”
  • include_comparisonboolean“Include tier comparison table (default true)”
• self-host-breakeven-calculator

  “Use when a user is deciding between API usage and self-hosted GPU inference at a given volume.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  6 inputs, 1 required

  • monthly_tokensnumberrequired“Monthly output token volume”
  • api_cost_per_1m_outnumber“Current API output cost per 1M tokens”
  • gpu_providerstring“GPU provider (e.g.”
  • gpu_typestring“GPU type (e.g.”
  • utilization_pctnumber“Expected GPU utilization % (default 60)”
  • operational_overhead_pctnumber“Ops overhead % on GPU cost (default 40)”
• context-window-planner

  “Use when a user needs to know whether a document plus prompt plus output fits within a model's context window, or wants a strategy recommendation (truncate/summarize/rag/chunk).”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  5 inputs, 2 required

  • doc_tokensnumberrequired“Primary document/content tokens”
  • overhead_tokensnumber“System prompt + few-shot + history (default 2000)”
  • expected_out_tokensnumber“Reserved output budget (default 1000)”
  • model_context_windownumberrequired“Target model context size”
  • strategy_hintstring“Preferred strategy (optional)”
• quantization-calculator

  “Use when a user is planning to quantize an LLM to fit on smaller hardware.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  5 inputs, 1 required

  • params_billionsnumberrequired“Model parameter count in billions”
  • precision_fromstring“Starting precision (default bf16)”
  • precision_tostring“Target precision (default int4)”
  • kv_cache_tokensnumber“Max KV cache tokens (default 8192)”
  • batch_sizenumber“Serving batch size (default 1)”
• fine-tune-roi-calculator

  “Use when a user is considering fine-tuning vs prompt engineering.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  6 inputs, 2 required

  • train_tokensnumberrequired“Training tokens (dataset × epochs)”
  • train_cost_per_1mnumber“Training cost per 1M tokens”
  • base_inference_cost_1mnumber“Baseline API output cost per 1M”
  • finetuned_inference_cost_1mnumber“Fine-tuned inference cost per 1M”
  • monthly_inference_tokensnumberrequired“Expected monthly inference volume (output tokens)”
  • prompt_reduction_pctnumber“Prompt size reduction % from eliminating few-shot (default 0)”
• eval-cost-calculator

  “Use when a user needs to budget an LLM evaluation run.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  10 inputs, 1 required

  • samplesnumberrequired“Number of eval samples”
  • modelsnumber“Candidate models (default 1)”
  • trials_per_samplenumber“Repeats per sample (default 1)”
  • avg_tokens_innumber“Avg input tokens per sample (default 2000)”
  • avg_tokens_outnumber“Avg output tokens per sample (default 500)”
  • providerstring“Eval provider (default anthropic)”
  • modelstring“Eval model (default claude-sonnet-4-6)”
  • judge_enabledboolean“Enable LLM-as-judge second pass (default false)”
  • judge_tokens_innumber“Judge input tokens (default 1500)”
  • judge_tokens_outnumber“Judge output tokens (default 200)”
• observability-cost-calculator

  “Use when a user needs to budget LLM observability tooling.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  4 inputs, 2 required

  • requests_per_daynumberrequired“Average daily LLM requests”
  • avg_log_bytesnumber“Avg payload bytes per traced request (default 4096)”
  • retention_daysnumber“Retention in days (default 30)”
  • providerstringrequired“Observability provider”
• rag-pipeline-cost-calculator

  “Use when a user needs end-to-end RAG cost estimation (embedding + vector store + generation).”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  11 inputs, 2 required

  • queries_per_daynumberrequired“User query volume per day”
  • corpus_tokensnumberrequired“Indexed corpus size in tokens”
  • chunk_size_tokensnumber“Avg chunk size (default 512)”
  • chunks_retrievednumber“Chunks per query (default 5)”
  • embedding_providerstring“Embedding model”
  • vector_storestring“Vector store”
  • generator_providerstring“Generator LLM provider (default anthropic)”
  • generator_modelstring“Generator model (default claude-sonnet-4-6)”
  • question_tokensnumber“Avg question tokens (default 100)”
  • answer_tokensnumber“Avg answer tokens (default 400)”
  • reindex_fraction_per_monthnumber“Fraction of corpus re-embedded per month (default 0.1)”
• automation-cost-calculator

  “Use for generic or non-agent recurring workflow platform billing across Zapier task billing, Make credit billing, and n8n execution billing.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  3 inputs, 1 required

  • workflow_runs_per_monthnumberrequired“Workflow/scenario runs per month”
  • billable_steps_per_runnumber“Billable actions/modules per run (default 5)”
  • make_modules_per_runnumber“Make module actions per scenario run; defaults to billable_steps_per_run”
• agent-workflow-cost-calculator

  “Use for automation-platform/iPaaS cost of an AI agent workflow, including app-action fan-out and MCP tool-call accounting, explicitly excluding LLM token spend.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  4 inputs, 1 required

  • agent_runs_per_monthnumberrequired“Agent workflow runs per month”
  • app_actions_per_runnumber“Downstream app actions per agent run (default 3)”
  • mcp_tool_calls_per_runnumber“MCP tool calls per agent run (default 1)”
  • make_modules_per_runnumber“Make modules per run; defaults to app actions + MCP calls”
• agent-loop-cost-calculator

  “Use for LLM inference/reasoning cost per successful multi-step agent task, including failure overhead and context growth across turns.”

  Publisher declares: read-only yes · destructive no · idempotent yes · open-world no

  9 inputs, 1 required

  • tasks_per_monthnumberrequired“Tasks attempted per month”
  • steps_per_tasknumber“Avg reasoning steps per task (default 5)”
  • tool_calls_per_stepnumber“Avg tool invocations per step (default 2)”
  • avg_tokens_in_per_turnnumber“Avg input tokens per turn (default 3000)”
  • avg_tokens_out_per_turnnumber“Avg output tokens per turn (default 400)”
  • success_ratenumber“Task success rate 0-1 (default 0.7)”
  • providerstring“LLM provider (default anthropic)”
  • modelstring“LLM model (default claude-sonnet-4-6)”
  • context_growth_factornumber“Multiplier on input tokens as conversation grows (default 1.4)”

Each quoted sentence is the first sentence of the publisher’s own description. The kind of tool and the “changes data” mark are our AI model’s reading of that description, not something we tested: two differently worded runs had to agree, otherwise no label is shown, and nobody has checked them by hand. A tool without the “changes data” mark is described as only reading, searching or calculating, or the two runs did not agree. No tool was called. Read the publisher’s full description before relying on it. “Publisher declares” lists the hints the server itself sends with the tool (MCP annotations: read-only, destructive, idempotent, open-world); they are the publisher's claims, not checked by us. Inputs are the parameters in the tool's own input schema: name, declared type, whether it is required, and the first sentence of its description. Read 2026-10-23.

The registry verifies a package once, when it is published, and does not re-check it. A difference below is about the listing, not a judgement of the publisher.