Keep the list current: six new projects and maintenance tooling

.github/ISSUE_TEMPLATE/add-project.yml

@@ -0,0 +1,73 @@
+name: Add a project
+description: Suggest a tool for the awesome list
+title: "Add [Project Name]"
+labels: ["addition"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thanks for helping curate Awesome Token Mining! Please read [CONTRIBUTING.md](../CONTRIBUTING.md) before submitting.
+
+  - type: input
+    id: project-name
+    attributes:
+      label: Project name
+      placeholder: e.g. LMCache
+    validations:
+      required: true
+
+  - type: input
+    id: project-url
+    attributes:
+      label: Canonical URL
+      description: GitHub repo or official docs
+      placeholder: https://github.com/org/repo
+    validations:
+      required: true
+
+  - type: dropdown
+    id: category
+    attributes:
+      label: Suggested category
+      options:
+        - Self-Hosting
+        - vLLM & High-Throughput Inference
+        - Quantization
+        - KV Cache & Speculative Decoding
+        - Token Compression
+        - Semantic & Response Caching
+        - Gateways & Model Routing
+        - Observability & Cost Tracking
+        - Context & Prompt Optimization
+        - New category (describe below)
+    validations:
+      required: true
+
+  - type: textarea
+    id: description
+    attributes:
+      label: One-line description
+      description: What it does and how it saves tokens or money
+      placeholder: "[Project] - Does X, saving Y% on Z workloads."
+    validations:
+      required: true
+
+  - type: textarea
+    id: cost-angle
+    attributes:
+      label: Cost-saving angle
+      description: How does this concretely reduce token spend, inference cost, or API bills?
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: checklist
+    attributes:
+      label: Checklist
+      options:
+        - label: The project is open source or has a meaningful free tier
+          required: true
+        - label: The project has had commits in the last 12 months
+          required: true
+        - label: I searched the README and this is not a duplicate
+          required: true

.github/pull_request_template.md

@@ -0,0 +1,17 @@
+## Summary
+
+<!-- What project(s) are you adding or updating, and why do they belong here? -->
+
+## Checklist
+
+- [ ] Link works and points to the canonical source
+- [ ] Entry is in alphabetical order within its section
+- [ ] Description is one line and explains the cost-saving value
+- [ ] No duplicate entries (searched the README)
+- [ ] Read [CONTRIBUTING.md](../CONTRIBUTING.md)
+
+## Entry preview
+
+```markdown
+<!-- Paste the exact line you're adding -->
+```

.github/workflows/link-check.yml

@@ -0,0 +1,28 @@
+name: Link Check
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+  schedule:
+    # Weekly on Mondays at 06:00 UTC
+    - cron: "0 6 * * 1"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  lychee:
+    name: Check links
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Run lychee
+        uses: lycheeverse/lychee-action@v2
+        with:
+          args: --verbose --no-progress README.md CONTRIBUTING.md CHANGELOG.md
+          fail: true

CHANGELOG.md

@@ -0,0 +1,18 @@
+# Changelog
+
+All notable updates to this list are documented here.
+
+## 2026-07-07
+
+### Added
+
+- **KV Cache & Speculative Decoding:** [LMCache](https://github.com/LMCache/LMCache), [Mooncake](https://github.com/kvcache-ai/Mooncake)
+- **Semantic & Response Caching:** [FerroCache](https://github.com/nickleodoen/ferrocache), [semcache](https://github.com/sensoris/semcache)
+- **Gateways & Model Routing:** [Bifrost](https://github.com/maximhq/bifrost), [UncommonRoute](https://github.com/CommonstackAI/UncommonRoute)
+
+### Maintenance
+
+- Fixed broken links for ModelCache, RedisVL, Martian, and Narev
+- Removed TokenForge (repository no longer available)
+- Added weekly link-check GitHub Action
+- Added issue and pull request templates for community contributions

CONTRIBUTING.md

@@ -33,7 +33,7 @@ Add projects that help people:
    ```
 
 4. Keep descriptions factual. Cite measurable claims when you can (e.g. "up to 20× compression" from the project's own benchmarks).
-5. Open a PR with a clear title like `Add LiteLLM to Gateways section`.
+5. Open a PR with a clear title like `Add LiteLLM to Gateways section`, or file an [issue using the add-project template](.github/ISSUE_TEMPLATE/add-project.yml).
 
 ## Pull request checklist
 

README.md

@@ -15,7 +15,7 @@
   <a href="https://awesome.re"><img src="https://awesome.re/badge.svg" alt="Awesome"></a>
   <img src="https://img.shields.io/github/stars/narevai/awesome-tokenminning?style=flat-square&color=111111&label=stars" alt="Stars">
   <img src="https://img.shields.io/badge/categories-10-111111?style=flat-square" alt="10 categories">
-  <img src="https://img.shields.io/badge/tools-50%2B-111111?style=flat-square" alt="50+ tools">
+  <img src="https://img.shields.io/badge/tools-55%2B-111111?style=flat-square" alt="55+ tools">
   <img src="https://img.shields.io/badge/license-CC0-111111?style=flat-square" alt="CC0 license">
 </p>
 
@@ -68,6 +68,8 @@ Attack the two biggest inference bottlenecks: KV cache memory on long contexts,
 
 - [EAGLE](https://github.com/SafeAILab/EAGLE) - Speculative decoding via feature extrapolation. A small draft head proposes multiple tokens; the main model verifies in parallel for ~2–3× faster generation with provably identical output distribution.
 - [kvpress](https://github.com/NVIDIA/kvpress) - NVIDIA's unified library of KV cache compression methods (SnapKV, H2O, Finch, and more). Drop-in Hugging Face pipeline for long-context workloads where cache size dominates memory.
+- [LMCache](https://github.com/LMCache/LMCache) - Distributed KV cache layer for vLLM and SGLang. Shares prefix KV across instances and tiers so repeated long contexts skip full prefill — lower TTFT and more throughput per GPU.
+- [Mooncake](https://github.com/kvcache-ai/Mooncake) - KVCache-centric disaggregated serving platform (powers Kimi). Pools and transfers KV across prefill/decode nodes via RDMA — up to 75% more requests at the same SLO on production workloads.
 - [Medusa](https://github.com/FasterDecoding/Medusa) - Adds lightweight prediction heads to a frozen base model to speculate multiple tokens per step. No separate draft model required.
 - [SnapKV](https://github.com/FasterDecoding/SnapKV) - Fine-tuning-free KV cache compression. Observes attention patterns in a prompt window, keeps only clustered important positions per head — up to 8× memory savings on 16K+ contexts.
 
@@ -82,31 +84,34 @@ Shrink prompts and outputs before they hit the bill. These tools target the toke
 - [LongLLMLingua](https://github.com/microsoft/LLMLingua) - Long-context variant that mitigates "lost in the middle" while compressing RAG prompts to a fraction of their size.
 - [LLMLingua-2](https://github.com/microsoft/LLMLingua) - Faster, task-agnostic prompt compression using a distilled model for near-real-time use.
 - [Ponytail](https://github.com/DietrichGebert/ponytail) - Agent skill that stops over-engineering via a YAGNI ladder: stdlib, native platform, and one-liners before custom code. Agentic benchmark: ~54% less code, ~22% fewer tokens, ~20% lower cost on real Claude Code sessions.
-- [TokenForge](https://github.com/Manavarya09/tokenforge) - Full-stack Rust optimizer for code, CLI output, conversation history, JSON, and MCP schemas. AST-aware folding with lossless SQLite-backed reversibility.
 - [Tokenless](https://github.com/TokenFleet-AI/tokenless) - Rust toolkit for schema compression, differential responses, TOON encoding, and command-output rewriting. Targets 60–90% savings on agent tool-call loops.
 
 ## Semantic & Response Caching
 
 Skip the LLM entirely when a similar question was already answered, or reuse provider-side prefix caches.
 
+- [FerroCache](https://github.com/nickleodoen/ferrocache) - Standalone Rust semantic cache service with fleet-wide sharing, MCP support, and drop-in OpenAI/Anthropic SDK wrappers. Survives app restarts — skip redundant LLM calls across your whole stack.
 - [GPTCache](https://github.com/zilliztech/GPTCache) - Semantic cache for LLM apps. Vector similarity matching returns cached responses for equivalent queries — up to 10× cost reduction on hit.
 - [LiteLLM](https://github.com/BerriAI/litellm) - Universal LLM gateway with in-memory, Redis, S3, and semantic caching backends. One integration for caching across providers.
-- [RedisVL](https://github.com/redis/redis-vl) - Redis vector library for building semantic caches with sub-millisecond lookups at scale.
-- [ModelCache](https://github.com/shibing624/ModelCache) - Multi-level semantic cache with embedding similarity and TTL management for production LLM apps.
+- [ModelCache](https://github.com/codefuse-ai/ModelCache) - Multi-level semantic cache with embedding similarity and TTL management for production LLM apps.
 - [OpenAI Prompt Caching Cookbook](https://github.com/openai/openai-cookbook/blob/main/examples/Prompt_Caching_201.ipynb) - Practical guide to structuring prompts for up to 90% off cached input tokens on OpenAI models.
+- [RedisVL](https://github.com/redis/redis-vl-python) - Redis vector library for building semantic caches with sub-millisecond lookups at scale.
+- [semcache](https://github.com/sensoris/semcache) - Rust HTTP proxy semantic cache for OpenAI, Anthropic, and Gemini. Drop-in caching with Prometheus metrics and a built-in dashboard — no app code changes required.
 
 ## Gateways & Model Routing
 
 Route easy tasks to cheap models and hard tasks to capable ones. Enforce budgets before spend happens.
 
+- [Bifrost](https://github.com/maximhq/bifrost) - High-performance AI gateway with semantic caching, budget limits, and automatic failover across 23+ providers. OpenAI-compatible drop-in with sub-15µs routing overhead.
 - [Conduit](https://github.com/ashita-ai/conduit) - ML-powered router using Thompson Sampling bandits. Learns which model handles each query type best, balancing cost, quality, and latency from live traffic.
 - [LiteLLM Proxy](https://github.com/BerriAI/litellm) - Production proxy with budget limits, rate limiting, load balancing, and fallback chains across 100+ models.
 - [LLMRouter](https://github.com/ulab-uiuc/llmrouter) - Research-grade routing library with 16+ strategies (KNN, MLP, Elo, graph-based, BERT routers). Unified CLI for training cost-aware routers on benchmark data.
-- [Martian](https://github.com/withmartian/martian) - Model router that picks the cheapest model that can handle each request.
+- [Martian](https://github.com/withmartian/martian-sdk-python) - Python SDK for Martian's LLM gateway and routers. Programmatically manage cost-aware routing across 200+ models via a unified API.
 - [OpenRouter](https://openrouter.ai/) - Unified API across providers with automatic routing to cheapest available model for a given capability tier.
 - [ParetoBandit](https://github.com/ParetoBandit/ParetoBandit) - Cost-aware contextual bandit router with online budget pacing. Adapts when model prices or quality shift — routing decisions in microseconds on CPU.
 - [Portkey](https://github.com/Portkey-AI/gateway) - AI gateway with caching, retries, load balancing, and observability hooks for multi-provider setups.
 - [RouteLLM](https://github.com/lm-sys/RouteLLM) - ML-based router that sends queries to strong or weak models based on difficulty. Up to 85% cost reduction with minimal quality loss.
+- [UncommonRoute](https://github.com/CommonstackAI/UncommonRoute) - Local OpenAI-compatible proxy that routes each agent step independently. Held-out SWE-bench: matched task quality at 53% lower API cost vs frontier-only.
 
 ## Observability & Cost Tracking
 
@@ -115,7 +120,7 @@ You can't optimize what you can't measure. These tools attribute token spend to
 - [CostPilot](https://github.com/aryanjp1/costpilot) - Self-hosted LLM cost dashboard with a 3-line Python SDK. Tracks spend, tokens, and latency per model/feature with forecasting and savings recommendations.
 - [Helicone](https://github.com/Helicone/helicone) - LLM observability proxy that logs every request with latency, cost, and cache hit metrics.
 - [llmwatch](https://github.com/DanMeon/llmwatch) - Lightweight SDK instrumentation for OpenAI, Anthropic, Google, and more. Tags costs by feature/user with bundled pricing for 1000+ models — no proxy required.
-- [Narev](https://github.com/narev-ai/narev) - AI billing and cost attribution SDK. Pin live pricing, calculate per-request COGS, and wire usage-based billing.
+- [Narev](https://github.com/narevai/ai-billing) - Open-source Vercel AI SDK middleware for usage-based billing. Captures token usage, resolves live model prices, and routes billing events to Stripe, Polar, or Lago.
 - [Shekel](https://github.com/arieradle/shekel) - Budget enforcement and cost tracking for agentic systems. One-line Python integration with OpenTelemetry metrics, circuit breakers, and per-tool spend limits.
 
 ## Context & Prompt Optimization
@@ -135,7 +140,7 @@ Engineering patterns and libraries that reduce context bloat without adding new
 
 ## Contributing
 
-Contributions welcome. Please read [CONTRIBUTING.md](CONTRIBUTING.md) before opening a PR.
+Contributions welcome. Please read [CONTRIBUTING.md](CONTRIBUTING.md) before opening a PR. See [CHANGELOG.md](CHANGELOG.md) for recent updates.
 
 **Quick rules:**