feat: add semtree, semstore, and trimcp to projects

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: Strawbang

src/data/projects.ts

@@ -38,4 +38,49 @@ Works out of the box with Claude Code, Cursor, Windsurf, and GitHub Copilot. One
     language: 'Rust',
     hasReleases: true,
   },
+  {
+    slug: 'semtree',
+    name: 'semtree',
+    description: 'Semantic code intelligence for Rust — tree-sitter parsing, embeddings, and RAG for multi-language codebases.',
+    longDescription: `semtree turns your codebase into a searchable semantic index. It parses source files with tree-sitter, extracts meaningful chunks (functions, structs, modules), embeds them locally with fastembed, and stores them in a vector index powered by usearch.
+
+The result: lightning-fast semantic search over your code, no cloud required. Use it as a CLI to index and query any codebase, or as a library to build RAG pipelines that inject relevant context into LLM prompts.
+
+Supports Rust, Python, JavaScript, TypeScript, and Go out of the box.`,
+    github: 'https://github.com/rustkit-ai/semtree',
+    crates: 'https://crates.io/crates/semtree-core',
+    install: 'cargo install semtree',
+    language: 'Rust',
+    hasReleases: true,
+  },
+  {
+    slug: 'semstore',
+    name: 'semstore',
+    description: 'Local semantic search for Rust applications — store text, search by meaning, no cloud required.',
+    longDescription: `semstore is a lightweight semantic store for Rust. Drop it into any application to add meaning-based search: store text entries, search by semantic similarity, get back the most relevant results.
+
+Powered by fastembed for local embeddings (BGE-Small, ~23 MB, no API key needed) and usearch for fast approximate nearest-neighbour search. Backed by SQLite so data persists across restarts.
+
+Bring your own embedder via the Embedder trait if you need a different model or a remote API.`,
+    github: 'https://github.com/rustkit-ai/semstore',
+    crates: 'https://crates.io/crates/semstore',
+    install: null,
+    language: 'Rust',
+    hasReleases: true,
+  },
+  {
+    slug: 'trimcp',
+    name: 'trimcp',
+    description: 'MCP proxy that reduces LLM token costs by 60–90% through compression and caching.',
+    longDescription: `trimcp sits between your LLM client and upstream MCP servers, compressing and caching tool outputs before they reach the model.
+
+It strips ANSI codes, minifies JSON, removes comments, deduplicates repeated lines, and collapses whitespace — all without losing meaning. The result is 60–90% fewer tokens per MCP tool response, lower API costs, and faster completions.
+
+Works transparently with any MCP-compatible client (Claude, Cursor, Windsurf…). Zero config to get started: point it at your existing MCP servers and it handles the rest.`,
+    github: 'https://github.com/rustkit-ai/trimcp',
+    crates: 'https://crates.io/crates/trimcp',
+    install: 'cargo install trimcp',
+    language: 'Rust',
+    hasReleases: true,
+  },
 ];