docs: promote the acknowledgements refresh and the reindex-warning removal

README.md

@@ -29,16 +29,6 @@ Or open `http://localhost:21847/dashboard` in your browser:
 
 <img src="site/public/img/dashboard-home.png" alt="The cix dashboard — server status at a glance and a guided “Connect Claude Code to cix” onboarding with copy-paste commands" width="900">
 
-> [!IMPORTANT]
-> **Reindex after upgrading the server.** Until the parsing/chunking/embedding
-> pipeline stabilizes, an upgrade can change how code is embedded. A reindex
-> brings every project onto the new pipeline; within a version, search is
-> consistent once reindexed.
->
-> The 0.13.0 vector-store change is the exception: it moves the vectors you
-> already have into SQLite by itself, on first boot, without re-embedding
-> anything ([`doc/VECTORSTORE.md`](doc/VECTORSTORE.md)).
-
 ---
 
 ## Why
@@ -83,7 +73,7 @@ Grep and fuzzy file search work fine for small projects. At scale they break dow
 │  └── embedded React dashboard + Swagger UI                      │
 │                                                                 │
 │  Indexing pipeline                                              │
-│  ├── tree-sitter/wasm (AST chunking, 30+ langs)  (wazero)       │
+│  ├── tree-sitter/wasm (AST chunking, 31 langs)   (wazero)       │
 │  ├── embedding provider (local llama.cpp / Voyage / OpenAI)     │
 │  ├── SQLite vector store (float32 BLOBs, streamed cosine scan)  │
 │  └── SQLite FTS5 mirror (BM25) + metadata (modernc/sqlite)      │
@@ -97,6 +87,8 @@ Grep and fuzzy file search work fine for small projects. At scale they break dow
 
 Pure-Go static binary; CUDA-image variants add a CUDA runtime layer for GPU embeddings. Workspace clones live in `<data-dir>/repos/`.
 
+**Why vectors live in SQLite.** Through v0.12.x cix used [chromem-go](https://github.com/philippgille/chromem-go), an in-memory vector database. It decodes every document of every collection into the heap at startup and never evicts, so memory was proportional to the index rather than to the work: a real 312k-document index cost **2.2 GB resident while idle** and 47 seconds before it could answer anything. v0.13.0 replaced it with a SQLite store that keeps embeddings as float32 BLOBs and scans them per query — the same index now idles at **tens of megabytes** and answers about a millisecond after boot. The trade is search latency, roughly 4× higher and far less sensitive to how many results you ask for. Existing indexes are imported automatically on first boot; nothing is re-embedded. Numbers, layout and the migration: [`doc/VECTORSTORE.md`](doc/VECTORSTORE.md).
+
 ---
 
 ## Quick Start
@@ -346,15 +338,8 @@ projects and teams that make it possible:
 
 **Indexing & storage**
 - [tree-sitter](https://tree-sitter.github.io/tree-sitter/) — AST-aware
-  chunking across 30+ languages, run via
+  chunking across 31 languages, run via
   [wazero](https://github.com/tetratelabs/wazero) (pure-Go WASM runtime).
-- [gotreesitter](https://github.com/odvcencio/gotreesitter) — the Go
-  tree-sitter binding cix's AST chunking first grew from; thank you for the
-  head start.
-- [chromem-go](https://github.com/philippgille/chromem-go) — the
-  embedded vector store cix shipped through v0.12.x, and the model its
-  collection semantics still follow. Now kept only to read a pre-0.13
-  index during the one-time import.
 - [modernc.org/sqlite](https://gitlab.com/cznic/sqlite) — cgo-free
   SQLite for project metadata, symbols, the FTS5/BM25 mirror, and (since
   v0.13.0) the vectors themselves.
@@ -367,6 +352,8 @@ projects and teams that make it possible:
   [oapi-codegen](https://github.com/oapi-codegen/oapi-codegen) —
   OpenAPI-as-source-of-truth codegen for the Go interface and TypeScript
   dashboard types.
+- [gronx](https://github.com/adhocore/gronx) — the crontab expressions
+  behind scheduled database maintenance.
 - [brotli](https://github.com/andybalholm/brotli) and the
   [Go](https://go.dev/) standard library and `golang.org/x` ecosystem.
 
@@ -382,19 +369,34 @@ projects and teams that make it possible:
 - [notify](https://github.com/rjeczalik/notify) — cross-platform filesystem
   watching for the index-on-change watcher.
 - [koanf](https://github.com/knadh/koanf) — layered configuration
-  (flags → env → `~/.cix/config.yaml`).
+  (flags → env → `~/.cix/config.yaml`), with
+  [validator](https://github.com/go-playground/validator) checking what
+  lands there.
+- [go-gitignore](https://github.com/sabhiram/go-gitignore) — `.cixignore`
+  and `.gitignore` matching, so the watcher and the indexer agree with git
+  about what is source.
 
 **Dashboard (web UI)**
 - [React](https://react.dev/) + [Vite](https://vitejs.dev/) — the embedded
-  dashboard served at `/dashboard`.
+  dashboard served at `/dashboard`, routed by
+  [React Router](https://reactrouter.com/).
 - [Radix UI](https://www.radix-ui.com/) + [Tailwind CSS](https://tailwindcss.com/)
   — accessible component primitives and styling (the shadcn/ui pattern).
 - [TanStack Query](https://tanstack.com/query) — server-state and data
   fetching.
 - [openapi-typescript](https://github.com/openapi-ts/openapi-typescript) —
   generates the dashboard's API types from the OpenAPI spec.
-- [lucide](https://lucide.dev/) and [sonner](https://github.com/emilkowalski/sonner)
-  — icons and toast notifications.
+- [sonner](https://github.com/emilkowalski/sonner) — toast notifications.
+
+**No longer in the stack, still owed thanks**
+- [chromem-go](https://github.com/philippgille/chromem-go) — the embedded
+  vector store cix shipped through v0.12.x, and the model its collection
+  semantics still follow. It is why the early versions worked at all; it
+  is kept as a dependency only to read a pre-0.13 index during the
+  one-time import.
+- [gotreesitter](https://github.com/odvcencio/gotreesitter) — the Go
+  tree-sitter binding cix's AST chunking first grew from, before the move
+  to WASM grammars on wazero. Thank you for the head start.
 
 Full dependency lists with versions live in
 [`server/go.mod`](server/go.mod), [`cli/go.mod`](cli/go.mod), and

doc/LANGUAGES.md

@@ -1,42 +1,42 @@
 # Supported languages
 
-cix uses tree-sitter (via `github.com/odvcencio/gotreesitter`) to extract semantic chunks (functions, classes, methods, types) from source code. Files in unsupported languages still get indexed via a sliding-window fallback — they're searchable, just without per-symbol granularity.
+cix uses tree-sitter to extract semantic chunks (functions, classes, methods, types) from source code. The grammars are compiled to WebAssembly and run in [wazero](https://github.com/tetratelabs/wazero), a pure-Go WASM runtime — so the server stays a cgo-free static binary and a runaway grammar hits a memory limit inside the sandbox instead of the process heap (`server/internal/chunker/tswasm`). Files in unsupported languages still get indexed via a sliding-window fallback — they're searchable, just without per-symbol granularity.
 
 ## Default language set (31)
 
-| ID | gotreesitter factory | Function | Class | Method | Type |
-|---|---|:-:|:-:|:-:|:-:|
-| `python` | `PythonLanguage` | ✓ | ✓ | | |
-| `typescript` | `TypescriptLanguage` | ✓ | ✓ | ✓ | ✓ |
-| `tsx` | `TsxLanguage` | ✓ | ✓ | ✓ | ✓ |
-| `javascript` | `JavascriptLanguage` | ✓ | ✓ | ✓ | |
-| `go` | `GoLanguage` | ✓ | | ✓ | ✓ |
-| `rust` | `RustLanguage` | ✓ | ✓ | | ✓ |
-| `java` | `JavaLanguage` | ✓ | ✓ | | ✓ |
-| `c` | `CLanguage` | ✓ | ✓ | | ✓ |
-| `cpp` | `CppLanguage` | ✓ | ✓ | | ✓ |
-| `c_sharp` | `CSharpLanguage` | ✓ | ✓ | ✓ | ✓ |
-| `ruby` | `RubyLanguage` | ✓ | ✓ | | |
-| `php` | `PhpLanguage` | ✓ | ✓ | ✓ | ✓ |
-| `swift` | `SwiftLanguage` | ✓ | ✓ | | ✓ |
-| `kotlin` | `KotlinLanguage` | ✓ | ✓ | | |
-| `scala` | `ScalaLanguage` | ✓ | ✓ | | ✓ |
-| `bash` | `BashLanguage` | ✓ | | | |
-| `lua` | `LuaLanguage` | ✓ | | | |
-| `dart` | `DartLanguage` | ✓ | ✓ | ✓ | ✓ |
-| `r` | `RLanguage` | ✓ | | | |
-| `objc` | `ObjcLanguage` | ✓ | ✓ | ✓ | ✓ |
-| `html` | `HtmlLanguage` | | | | ✓ |
-| `css` | `CssLanguage` | | ✓ | | |
-| `scss` | `ScssLanguage` | ✓ | ✓ | | |
-| `sql` | `SqlLanguage` | ✓ | | | ✓ |
-| `markdown` | `MarkdownLanguage` | | | | ✓ |
-| `zig` | `ZigLanguage` | ✓ | ✓ | | |
-| `julia` | `JuliaLanguage` | ✓ | | | |
-| `fortran` | `FortranLanguage` | ✓ | ✓ | | |
-| `haskell` | `HaskellLanguage` | ✓ | | | ✓ |
-| `ocaml` | `OcamlLanguage` | ✓ | ✓ | | ✓ |
-| `solidity` | `SolidityLanguage` | ✓ | ✓ | | ✓ |
+| ID | Function | Class | Method | Type |
+|---|:-:|:-:|:-:|:-:|
+| `python` | ✓ | ✓ | | |
+| `typescript` | ✓ | ✓ | ✓ | ✓ |
+| `tsx` | ✓ | ✓ | ✓ | ✓ |
+| `javascript` | ✓ | ✓ | ✓ | |
+| `go` | ✓ | | ✓ | ✓ |
+| `rust` | ✓ | ✓ | | ✓ |
+| `java` | ✓ | ✓ | | ✓ |
+| `c` | ✓ | ✓ | | ✓ |
+| `cpp` | ✓ | ✓ | | ✓ |
+| `c_sharp` | ✓ | ✓ | ✓ | ✓ |
+| `ruby` | ✓ | ✓ | | |
+| `php` | ✓ | ✓ | ✓ | ✓ |
+| `swift` | ✓ | ✓ | | ✓ |
+| `kotlin` | ✓ | ✓ | | |
+| `scala` | ✓ | ✓ | | ✓ |
+| `bash` | ✓ | | | |
+| `lua` | ✓ | | | |
+| `dart` | ✓ | ✓ | ✓ | ✓ |
+| `r` | ✓ | | | |
+| `objc` | ✓ | ✓ | ✓ | ✓ |
+| `html` | | | | ✓ |
+| `css` | | ✓ | | |
+| `scss` | ✓ | ✓ | | |
+| `sql` | ✓ | | | ✓ |
+| `markdown` | | | | ✓ |
+| `zig` | ✓ | ✓ | | |
+| `julia` | ✓ | | | |
+| `fortran` | ✓ | ✓ | | |
+| `haskell` | ✓ | | | ✓ |
+| `ocaml` | ✓ | ✓ | | ✓ |
+| `solidity` | ✓ | ✓ | | ✓ |
 
 The exact AST node types per language live in `server/internal/chunker/chunker.go` (`defaultRegistry`). File-extension mapping lives in `server/internal/langdetect/langdetect.go`.
 
@@ -66,7 +66,7 @@ These produce sliding-window chunks. Adding semantic chunking is a one-map-entry
 
 `erlang, elixir, commonlisp, svelte, graphql, hcl (terraform), cmake, dockerfile, regex, xml, make`
 
-PRs welcome — verify node names with `gotreesitter`'s `cmd/tsquery` against a representative fixture before adding.
+PRs welcome — the node names in `defaultRegistry` must match what the grammar actually emits, and `chunker_test.go` walks the registry against the compiled grammars to catch a typo without needing a fixture file per language.
 
 ## How the chunker decides
 

doc/TEAM_DEPLOYMENT.md

@@ -251,10 +251,12 @@ docker compose pull
 docker compose up -d
 ```
 
-> **Reindex after a server upgrade.** Until the parsing/chunking/embedding
-> pipeline stabilizes, an upgrade can change how code is embedded. Trigger a
-> reindex (dashboard or `cix reindex`) so every project lands on the new
-> pipeline. Within a version, search is consistent once reindexed.
+> **An upgrade does not normally need a reindex.** Existing indexes keep
+> working, and 0.13's move of the vectors into SQLite imports them for you on
+> first boot without re-embedding anything. The cases that do call for one —
+> the embedding model changing, or a release that backfills new chunk-level
+> data — are listed in [`UPDATES.md`](UPDATES.md#3-reindex-after-an-upgrade),
+> and the dashboard flags affected projects itself.
 
 Server and CLI are released on **independent tag streams** (`server/vX.Y.Z`
 vs CLI tags) and are wire-compatible across a minor skew — you don't have to

doc/UPDATES.md

@@ -127,7 +127,7 @@ Upgrading the **server** can require a reindex in two cases:
    model (or you change it yourself via Dashboard → Server →
    Embedding model), every project becomes stale. The dashboard's
    drift indicator paints affected projects red with a "Stale model"
-   badge until you reindex. See README's *Drift indicator* section.
+   badge until you reindex. See [`DASHBOARD.md`](DASHBOARD.md#drift-indicator).
 2. **Schema migration adds chunk-level data.** Releases that backfill
    new chunk metadata (e.g. the FTS5 mirror introduced by `f00e3d3`)
    may prompt the dashboard to recommend a reindex on existing

server/dashboard/src/modules/projects/components/ProjectInfoCard.tsx

@@ -4,7 +4,7 @@ import { formatBytes } from '@/lib/formatBytes';
 import { formatDateTime, formatRelative } from '@/lib/formatDate';
 
 // Metadata that would otherwise need an SSH session: which embedding model
-// produced the vectors, where this project's SQLite and chromem-go state
+// produced the vectors, where this project's SQLite and vector-store state
 // lives, and how big both are. Storage fields are nullable — an
 // embeddings-disabled server has no resolvable paths, and "—" beats "0 B".
 //

server/internal/db/schema.go

@@ -417,7 +417,7 @@ CREATE INDEX IF NOT EXISTS idx_call_edges_caller ON call_edges(caller_symbol);
 CREATE INDEX IF NOT EXISTS idx_call_edges_callee ON call_edges(callee_symbol);
 
 -- PR14 dropped the workspaces communities/community_members tables.
--- Workspace search is now a weighted fan-out across per-project chromem
+-- Workspace search is now a weighted fan-out across per-project vector
 -- collections (no Louvain, no centroid index). migrateDropCommunities
 -- DROPs the tables on upgrade for installs that ran any of PR5..PR12.
 
@@ -441,8 +441,8 @@ CREATE INDEX IF NOT EXISTS idx_chunks_meta_project_file
 CREATE INDEX IF NOT EXISTS idx_chunks_meta_project
     ON chunks_meta(project_path);
 
--- chunks_fts is the BM25-searchable side, parallel to chromem-go's dense
--- vector store. Workspace search runs both in parallel per project then
+-- chunks_fts is the BM25-searchable side, parallel to the dense vector
+-- store. Workspace search runs both in parallel per project then
 -- fuses by RRF; project-relevance gating uses BM25 signal to drop repos
 -- that share no token with the query (the dense-only fan-out leaks
 -- semantically-distant repos as false positives).

server/internal/embeddings/provider/ollama/provider.go

@@ -110,9 +110,9 @@ func (p *Provider) ID() string {
 	return "ollama:" + p.cfg.Model
 }
 
-// Dimension returns 0 — the vector store infers dimension from the
-// first upsert (chromem-go behaviour) and CodeRankEmbed-Q8 reports
-// it on first call.
+// Dimension returns 0 — the vector store takes whatever width the first
+// upsert carries (each namespace holds one model's vectors, so they cannot
+// mix) and CodeRankEmbed-Q8 reports it on first call.
 func (p *Provider) Dimension() int { return 0 }
 
 // SupportsTokenize is true: llama-server exposes /tokenize.

server/internal/embeddings/provider/voyage/voyage.go

@@ -9,7 +9,7 @@
 //     (256/512/1024/2048). Part of Provider.ID() because changing it
 //     invalidates the existing index.
 //   - output_dtype: float|int8 (binary/ubinary are out of scope —
-//     chromem-go has no hamming search). For int8 the server returns
+//     the vector store has no hamming search). For int8 the server returns
 //     a list of integers per dimension; we dequantize to float32 in
 //     this package before returning vectors to the vector store.
 //
@@ -734,7 +734,7 @@ func (p *Provider) embed(ctx context.Context, texts []string, inputType string)
 // the approximate unit-norm float representation.
 //
 // This is the only place in the codebase that handles int8 quantized
-// embeddings; chromem-go and the search path both work exclusively
+// embeddings; the vector store and the search path both work exclusively
 // in float32.
 func dequantize(raw json.RawMessage, dtype string) ([]float32, error) {
 	switch dtype {

server/internal/httpapi/router.go

@@ -76,7 +76,7 @@ type Deps struct {
 	// server is started with CIX_EMBEDDINGS_ENABLED=false (e.g. in router
 	// tests). Phase 5 uses it for semantic search.
 	EmbeddingSvc EmbeddingsQuerier
-	// VectorStore is the chromem-go backed vector store (Phase 4). Nil-safe:
+	// VectorStore is the SQLite-backed vector store. Nil-safe:
 	// semantic search returns empty results when absent. Typed as the
 	// vectorstore.Interface so production can supply a *vectorstore.Holder
 	// (swappable on provider switch) while tests pass a raw *Store.

server/internal/httpapi/workspacesearch.go

@@ -119,14 +119,14 @@ type projectHits struct {
 // WorkspaceSearch — GET /api/v1/workspaces/{id}/search.
 //
 // Hybrid BM25+dense fan-out. Each project runs two queries in
-// parallel: dense (chromem cosine) and sparse (SQLite FTS5 BM25 over
+// parallel: dense (vector-store cosine) and sparse (SQLite FTS5 BM25 over
 // chunks_fts). Per project, the two ranked lists are fused via
 // Reciprocal Rank Fusion. Across projects, an α-blended candidacy
 // score (with per-query min-max normalization on both signals) plus
 // a relative threshold (`candidacy ≥ best × 0.4`) keeps the result
 // set focused on repos that actually share vocabulary or semantics
 // with the query — pure-dense fan-out leaked every workspace repo at
-// noise-level cosine similarity, since chromem returns the N nearest
+// noise-level cosine similarity, since a top-K scan returns the N nearest
 // vectors regardless of how far away "nearest" actually is.
 //
 // Observed pre-hybrid: in a workspace of N repos, the repos that
@@ -756,9 +756,9 @@ func projectLabel(projectPath string) string {
 	return projectPath
 }
 
-// round4 rounds f to 4 decimal places — matches the chunk-side
-// rounding chromem already applies, so scores in the response look
-// consistent across nested fields.
+// round4 rounds f to 4 decimal places — matches the chunk-side rounding the
+// vector store already applies, so scores in the response look consistent
+// across nested fields.
 func round4(f float32) float32 {
 	if math.IsNaN(float64(f)) {
 		return 0

server/internal/indexer/indexer.go

@@ -818,8 +818,9 @@ func (s *Service) ProcessFilesStreaming(
 	}
 
 	// ---- Stage 3: WRITE (serial, ordered) --------------------------------
-	// Vector-store + per-file DB writes run on this single goroutine: chromem
-	// is thread-safe but serialising keeps SQLite's WAL writer uncontended and
+	// Vector-store + per-file DB writes run on this single goroutine: the
+	// store is thread-safe, but serialising keeps SQLite's WAL writer
+	// uncontended and
 	// preserves deterministic progress-event ordering. Each write is local and
 	// sub-ms, so serialising costs nothing next to the (now parallel) embeds.
 	for _, p := range prep {
@@ -868,7 +869,7 @@ func (s *Service) ProcessFilesStreaming(
 			}
 		}
 
-		// Build chunksfts payload from the same chunks pushed to chromem.
+		// Build chunksfts payload from the same chunks pushed to the vector store.
 		ftsChunks := make([]chunksfts.Chunk, len(p.vsChunks))
 		for i, c := range p.vsChunks {
 			ftsChunks[i] = chunksfts.Chunk{

server/internal/maintenance/analyze.go

@@ -96,7 +96,7 @@ func (s *Service) Analyze(ctx context.Context) (*Analysis, error) {
 // scanState is the live picture every scanner reconciles against, read once so
 // five scanners do not issue the same queries five times.
 type scanState struct {
-	// liveCollections maps chromem collection name -> host_path, for every
+	// liveCollections maps vector-store collection name -> host_path, for every
 	// row currently in `projects`.
 	liveCollections map[string]string
 	// liveHashes is the set of projects.HashPath values currently in use.

server/internal/vectorstore/store.go

@@ -11,7 +11,8 @@
 // embedding namespace and a query streams the collection past a dot product
 // with a top-K heap. Opening is one file open (sub-millisecond), resident
 // memory is a couple of page-cache buffers per active query, and the cost is
-// search latency: roughly 3x chromem's on the same data.
+// search latency: roughly 4x chromem's on the same data — see the measured
+// table in doc/VECTORSTORE.md.
 //
 // # Frozen contracts
 //