From 87b4113bcbd53eacffeb8a94167cd7333140e57d Mon Sep 17 00:00:00 2001
From: RicardoAGL <ricardoangelr2@gmail.com>
Date: Wed, 13 May 2026 12:18:44 +0200
Subject: [PATCH 1/2] fix(lattice): normalize Windows path separators in
 section/ref file IDs

On Windows, path.relative() returns backslash-separated paths.
These backslashes were embedded raw into section IDs and ref file
fields, causing wiki-link resolution to fail with false broken-link
errors on Windows (e.g. 8 spurious errors in a clean project).

Fix: chain .replace(/\/g, '/') after the .md strip in both
parseSections() and extractRefs(). This is a no-op on Mac/Linux
where relative() already returns forward slashes.

Fixes broken `lat check` output on Windows.
---
 src/lattice.ts | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/lattice.ts b/src/lattice.ts
index 857a345..d207530 100644
--- a/src/lattice.ts
+++ b/src/lattice.ts
@@ -99,7 +99,7 @@ export function parseSections(
 ): Section[] {
   const tree = parse(content);
   const file = projectRoot
-    ? relative(projectRoot, filePath).replace(/\.md$/, '')
+    ? relative(projectRoot, filePath).replace(/\.md$/, '').replace(/\\/g, '/')
     : basename(filePath, '.md');
   const sectionFilePath = projectRoot
     ? relative(projectRoot, filePath)
@@ -619,7 +619,7 @@ export function extractRefs(
 ): Ref[] {
   const tree = parse(content);
   const file = projectRoot
-    ? relative(projectRoot, filePath).replace(/\.md$/, '')
+    ? relative(projectRoot, filePath).replace(/\.md$/, '').replace(/\\/g, '/')
     : basename(filePath, '.md');
   const refs: Ref[] = [];
 

From 7d8a0af709c7c2ffc232f64c75a04864c29b8476 Mon Sep 17 00:00:00 2001
From: RicardoAGL <ricardoangelr2@gmail.com>
Date: Wed, 13 May 2026 13:15:32 +0200
Subject: [PATCH 2/2] feat(search): add Ollama embedding provider
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Adds `ollama:<model>` key format to LAT_LLM_KEY, enabling local
embedding generation via Ollama without an external API key.

Usage:
  LAT_LLM_KEY=ollama:nomic-embed-text lat search "query"

Environment variables:
  OLLAMA_HOST           Ollama base URL (default: http://localhost:11434)
  LAT_OLLAMA_DIMENSIONS Override vector dimensions for unlisted models

Known dimensions are pre-configured for: nomic-embed-text (768),
mxbai-embed-large (1024), all-minilm (384), snowflake-arctic-embed (1024),
bge-large (1024), bge-base (768), bge-small (384).

For any other model, set LAT_OLLAMA_DIMENSIONS to the model's output size.
The Ollama /v1/embeddings endpoint is OpenAI-compatible — no auth header
is sent (Ollama runs locally and requires none).
---
 src/search/provider.ts | 33 ++++++++++++++++++++++++++++++++-
 1 file changed, 32 insertions(+), 1 deletion(-)

diff --git a/src/search/provider.ts b/src/search/provider.ts
index 16ee2b6..c838d72 100644
--- a/src/search/provider.ts
+++ b/src/search/provider.ts
@@ -6,6 +6,18 @@ export type EmbeddingProvider = {
   headers: (key: string) => Record<string, string>;
 };
 
+// Known dimensions for common Ollama embedding models.
+// Override with LAT_OLLAMA_DIMENSIONS for unlisted models.
+const OLLAMA_KNOWN_DIMENSIONS: Record<string, number> = {
+  'nomic-embed-text': 768,
+  'mxbai-embed-large': 1024,
+  'all-minilm': 384,
+  'snowflake-arctic-embed': 1024,
+  'bge-large': 1024,
+  'bge-base': 768,
+  'bge-small': 384,
+};
+
 const openai: EmbeddingProvider = {
   name: 'openai',
   apiBase: 'https://api.openai.com/v1',
@@ -39,6 +51,25 @@ export function detectProvider(key: string): EmbeddingProvider {
       headers: () => ({ 'Content-Type': 'application/json' }),
     };
   }
+  if (key.startsWith('ollama:')) {
+    const model = key.slice('ollama:'.length).trim();
+    if (!model)
+      throw new Error(
+        'ollama: key must include a model name, e.g. LAT_LLM_KEY=ollama:nomic-embed-text',
+      );
+    const host = process.env.OLLAMA_HOST ?? 'http://localhost:11434';
+    const dimOverride = process.env.LAT_OLLAMA_DIMENSIONS;
+    const dimensions = dimOverride
+      ? parseInt(dimOverride, 10)
+      : (OLLAMA_KNOWN_DIMENSIONS[model] ?? 768);
+    return {
+      name: 'ollama',
+      apiBase: `${host}/v1`,
+      model,
+      dimensions,
+      headers: () => ({ 'Content-Type': 'application/json' }),
+    };
+  }
   if (key.startsWith('sk-ant-')) {
     throw new Error(
       "Anthropic doesn't offer an embedding model. Set LAT_LLM_KEY to an OpenAI (sk-...) or Vercel AI Gateway (vck_...) key.",
@@ -47,6 +78,6 @@ export function detectProvider(key: string): EmbeddingProvider {
   if (key.startsWith('vck_')) return vercel;
   if (key.startsWith('sk-')) return openai;
   throw new Error(
-    `Unrecognized LAT_LLM_KEY prefix. Supported: OpenAI (sk-...), Vercel AI Gateway (vck_...).`,
+    `Unrecognized LAT_LLM_KEY prefix. Supported: OpenAI (sk-...), Vercel AI Gateway (vck_...), Ollama (ollama:<model>).`,
   );
 }