# Tokenminning > Tokenminning is a community wiki for understanding, measuring, and optimizing LLM token usage in production. Built by [Narev](https://narev.ai). Tokenminning is the deliberate counter-move to **tokenmaxxing**—maximizing raw AI usage without tying spend to outcomes. This site publishes the movement's engineering standards, conceptual guides, and practical techniques for controlling LLM costs: model routing, context hygiene, prompt caching, fiscal ceilings, and cost attribution. The [homepage](https://tokenminning.ai/) introduces tokenminning, points to the [Manifesto](https://tokenminning.ai/manifesto) and [Constitution](https://tokenminning.ai/constitution), and links into [Practice](https://tokenminning.ai/practice) for optimization playbooks. Start with [Concepts](https://tokenminning.ai/concepts) for token economics. ## Core documents - [Constitution](https://tokenminning.ai/constitution): Engineering law for production AI cost control — deployment gates, spending caps, and immutable metering ledgers - [Manifesto](https://tokenminning.ai/manifesto): Overview of the philosophical case for treating inference tokens as a scarce resource - [About](https://tokenminning.ai/about): Project purpose, what we publish, and how to contribute ## Constitution - [Article I: immutable metering](https://tokenminning.ai/constitution/article-i): Permanent, tamper-evident records of token usage - [Article II: the routing mandate](https://tokenminning.ai/constitution/article-ii): Match model tier to task complexity - [Article III: context window sovereignty](https://tokenminning.ai/constitution/article-iii): Control what enters the context window on every turn - [Article IV: the fiscal ceilings](https://tokenminning.ai/constitution/article-iv): Hard budget limits enforced in CI/CD and runtime - [Article V: prompt schema standards](https://tokenminning.ai/constitution/article-v): Structured, versioned prompt templates - [Article VI: the shadow stack prohibition](https://tokenminning.ai/constitution/article-vi): No unmonitored or unbudgeted model calls ## Manifesto - [Chapter I: the production gap](https://tokenminning.ai/manifesto/chapter-i): Why inference spend drifts before anyone notices - [Chapter II: tokenmaxxing](https://tokenminning.ai/manifesto/chapter-ii): The unnamed default of expanding token use - [Chapter III: tokenminning](https://tokenminning.ai/manifesto/chapter-iii): What the movement advocates - [Chapter IV: scale and unit economics](https://tokenminning.ai/manifesto/chapter-iv): Energy, finance, and measurement - [Chapter V: commitments](https://tokenminning.ai/manifesto/chapter-v): Closing pledges for practitioners ## Concepts - [Tokenminning](https://tokenminning.ai/concepts/tokenminning): Deliberately reducing LLM token consumption while preserving useful output - [Input and output tokens](https://tokenminning.ai/concepts/input-output-tokens): How providers bill for prompt and completion tokens - [Context inflation](https://tokenminning.ai/concepts/context-inflation): Why token counts grow faster than per-token prices - [Model selection](https://tokenminning.ai/concepts/model-selection): Routing and evaluation strategies for cost-quality tradeoffs ## Alternatives - [Alternatives](https://tokenminning.ai/alternatives): Tool and stack swaps compared by token economics — SEO/GEO hub for "alternative to X" queries - [Alternatives to Cursor](https://tokenminning.ai/alternatives/cursor): Windsurf, Copilot, Cline, Claude Code, Zed, and BYOK agents vs Cursor subscription and cache metering - [Alternatives to GitHub Copilot](https://tokenminning.ai/alternatives/github-copilot): Cursor, Cline, Windsurf, and terminal agents vs Copilot premium requests — per-tool breakdowns and token economics - [Alternatives to Ollama](https://tokenminning.ai/alternatives/ollama): vLLM, llama.cpp, TGI, and LocalAI vs Ollama dev pull-and-chat — GPU economics, hybrid patterns, per-stack breakdowns - [Alternatives to vLLM](https://tokenminning.ai/alternatives/vllm): Ollama, TGI, llama.cpp, and LocalAI vs vLLM production batching — when throughput ops are overkill ## Practice - [Where to start](https://tokenminning.ai/practice/where-to-start): Prioritized entry points for token optimization - [Prompt hygiene](https://tokenminning.ai/practice/prompt-hygiene): Trim system prompts and instructions without losing signal - [Prompt caching](https://tokenminning.ai/practice/prompt-caching): Reuse identical input to avoid re-billing - [Context hygiene](https://tokenminning.ai/practice/context-hygiene): Summarize, truncate, and bound agent history - [Model routing](https://tokenminning.ai/practice/model-routing): Route tasks to the cheapest model that meets quality bars - [Output and RAG](https://tokenminning.ai/practice/output-and-rag): Control completion length and retrieval payload size - [Semantic caching](https://tokenminning.ai/practice/semantic-caching): Cache responses for semantically similar queries - [Local inference](https://tokenminning.ai/practice/local-inference): Run Ollama, vLLM, or OpenAI-compatible servers instead of per-token APIs ## Self-hosting - [Self-hosting](https://tokenminning.ai/self-hosting): When on-prem inference beats cloud APIs — GPU metering, workload fit, and stack comparison - [Text generation inference](https://tokenminning.ai/self-hosting/text-generation-inference): Runtime comparison — Ollama, vLLM, llama.cpp, TGI, and LocalAI by workload profile - [Ollama](https://tokenminning.ai/self-hosting/ollama): Local dev inference, IDE integration, hybrid routing, and GPU time metering - [vLLM](https://tokenminning.ai/self-hosting/vllm): Production serving with continuous batching, GPU sizing, and observability - [llama.cpp](https://tokenminning.ai/self-hosting/llama-cpp): GGUF edge inference on CPU, Apple Silicon, and modest GPUs - [Text Generation Inference](https://tokenminning.ai/self-hosting/tgi): TGI inference engine — Hugging Face production serving for Kubernetes - [TGI server](https://tokenminning.ai/self-hosting/tgi/server): Run the Text Generation Inference server with Docker - [TGI API](https://tokenminning.ai/self-hosting/tgi/api): OpenAI-compatible API for Text Generation Inference - [TGI on Kubernetes](https://tokenminning.ai/self-hosting/tgi/kubernetes): Production K8s deploys for TGI - [TGI observability](https://tokenminning.ai/self-hosting/tgi/observability): Prometheus metrics and span tags for TGI - [TGI guardrails](https://tokenminning.ai/self-hosting/tgi/guardrails): Progressive enforcement for Text Generation Inference - [TGI support](https://tokenminning.ai/self-hosting/tgi/support): Enterprise support, Inference Endpoints, HF token management - [LocalAI](https://tokenminning.ai/self-hosting/localai): OpenAI-compatible gateway over multiple inference backends ## IDEs - [IDEs](https://tokenminning.ai/ides): Tokenminning guides for AI coding environments — tool comparison and per-editor optimization - [Cursor](https://tokenminning.ai/ides/cursor): Measure and reduce token burn in Cursor - [Windsurf](https://tokenminning.ai/ides/windsurf): Quota metering, SWE models, Cascade rules, and context pinning - [Devin](https://tokenminning.ai/ides/devin): ACUs, Adaptive routing, cloud sessions, and local agents - [OpenCode](https://tokenminning.ai/ides/opencode): `opencode stats`, Plan/Build routing, compaction, and MCP - [Claude Code](https://tokenminning.ai/ides/claude-code): Anthropic usage, CLAUDE.md, bash output, and CI caps - [Cline](https://tokenminning.ai/ides/cline): VS Code agent tasks, MCP, and auto-approve gates - [Aider](https://tokenminning.ai/ides/aider): Repo map tokens, architect mode, and git-scoped edits - [GitHub Copilot](https://tokenminning.ai/ides/github-copilot): Premium requests, chat vs inline vs agent mode - [Zed](https://tokenminning.ai/ides/zed): Assistant threads, model routing, and context attachment - [Trae](https://tokenminning.ai/ides/trae): Builder sessions, adaptive context, and quota guardrails - [OpenRouter](https://tokenminning.ai/ides/openrouter): BYOK gateway caps, model fallbacks, and Activity exports ## Optional - [AI Catalog](https://tokenminning.ai/.well-known/ai-catalog.json): Agentic Resource Discovery manifest - [Press](https://tokenminning.ai/press): Media coverage of the tokenminning movement - [GitHub repository](https://github.com/narevai/tokenminning-wiki): Issues and community feedback - [Sitemap](https://tokenminning.ai/sitemap.xml): Full list of indexable pages