We aim to be the SOTA in terms of deep research model and scaffold efficiency knowledge, and to share and use that knowledge to provide a meta deep research API that guarantees as good or better than current SOTA frameworks.
APIcostX is a text file based batch LLM workflow engine with GUI and API that sends requests to multiple models with different scaffolds, then evaluates and processes the result, for the purpose of generating a novel ground truth corpus. Efficiency information learned during execution is used to automatically reduce generation and evaluation calls on subsequent runs, continuously optimizing costs. Users can optimize API costs before execution using the pre-run cost calculator and our model/scaffold efficiency knowledgebase. APIcostX is the API cost multiplier: spend more at first to find the lowest-cost repeatable path that still meets the quality bar, then use that path to reduce API calls on future runs.
APIcostX offers GPT-R, AI-Q, MSAgents, and similar local deep research scaffolds as a service alongside scaffold-less passthrough API shape normalization service. APIcostX has no close peer for reasoning and web-search normalization across providers. Provider-side reasoning and grounding are strictly enforced for scaffold-less calls.
Visualizations make debugging workflows easy. Discovering and identifying issues with the model, scaffold, or input is fast because the problems jump off the page in full color.
Deep Research is our focus, but APIcostX is a general purpose LLM workflow engine. APIcostX can be used as an easy way to request a one-shot with reasoning and websearch, or to run full sophisticated recursive workflow pipelines, and everything in between.
APIcostX.com employs per-user, client-hashed password-derived-key encrypted databases. APIcostX aims to have zero knowledge at rest, therefore APIcostX has no ability to read user data of any kind at rest.
Each input file can produce a corresponding output file, and folders can act as batch input and output units. Massive, long-running unattended tasks are supported with concurrency, error handling, and retry management. Github can be used for Input and Output files.
Buy credits directly from apicostx.com or “BYOK” bring your own provider keys. Supported providers include OpenAI, Anthropic, Google, OpenRouter (Eva), and Tavily. APIcostX supports all available frontier provider-side deep research APIs.
APIcostX can evaluate existing documents, evaluate LLMs as generators, or as judges. APIcostX can compare direct model calls with deep research providers, scaffolds, and model/scaffold combinations.
Adjacent software: DeepResearchBench, GPT-Reseracher, LiteLLM, OpenRouter, Tavily, provider SDKs, middleware CLIs, and general LLM flowchart based pipeline tools.
Example Workflows
Translate 100 documents into another language and verify accuracy.
Generate a one-click new Wikipedia-style corpus using the full pipeline starting with Doc2Sourus.
Build an opinionated, novel, up-to-date RAG corpus for an OpenClaw agent.
Use daily news updates to refresh a RAG corpus or fine-tuning dataset and extend a model’s knowledge cutoff.
Evaluate existing documents, test for adherence to instructions, grade homework.