fix(ai): Ollama compatibility fixes for local-first usage (#1854 takeover)

Four of the five fixes from #1854, rebased onto current master:

1. budget-tracker: FREE_LOCAL_CHAT_PROVIDERS (ollama, llama-server)
   returns zero pricing for local chat providers so --max-cost bounded
   brainstorm/lsd no longer TX2 hard-fails on local inference. Sibling
   to FREE_LOCAL_EMBED_PROVIDERS / FREE_LOCAL_RERANK_PROVIDERS.
2. brainstorm/judges: coerce numeric idea ids (5 -> "Idea 05") before
   the string type check; some models (e.g. Kimi K2 via Ollama) emit
   numeric ids and previously had every idea silently rejected.
3. think: the buildGracefulMessage "no LLM available" text now points
   at chat_model / expansion_model / anthropic_api_key, not just the
   Anthropic key, since that sentinel can fire for any provider.
4. ollama recipe: add chat + expansion touchpoints (llama3.3, qwen3,
   deepseek-r1; supports_tools, zero cost) so recipes can route
   reasoning calls to Ollama.

The fifth fix from #1854 (scope the NO_ANTHROPIC_API_KEY warning to
the anthropic provider) was superseded on master by the #1698 probe
work: probeChatModel only returns 'unavailable' for anthropic-no-key,
so non-Anthropic providers already get MODEL_NOT_USABLE instead.

Co-authored-by: Luna <luna@starstudiolabs.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/core/ai/recipes/ollama.ts

@@ -33,6 +33,20 @@ export const ollama: Recipe = {
       // OLLAMA_NUM_PARALLEL config; no static cap to declare. v0.32 (#779).
       no_batch_cap: true,
     },
+    chat: {
+      models: ['llama3.3', 'qwen3', 'deepseek-r1'],
+      supports_tools: true,
+      supports_subagent_loop: false,
+      supports_prompt_cache: false,
+      cost_per_1m_input_usd: 0,
+      cost_per_1m_output_usd: 0,
+      price_last_verified: '2026-06-02',
+    },
+    expansion: {
+      models: ['llama3.3', 'qwen3', 'deepseek-r1'],
+      cost_per_1m_tokens_usd: 0,
+      price_last_verified: '2026-06-02',
+    },
   },
   setup_hint: 'Install Ollama from https://ollama.ai, then `ollama pull nomic-embed-text` and `ollama serve`.',
 };

src/core/brainstorm/judges.ts

@@ -361,6 +361,11 @@ function isAxisScoreInRange(n: unknown): n is number {
 function validateIdeaShape(raw: unknown): { id: string; scores: JudgeAxisScores; note: string } | null {
   if (typeof raw !== 'object' || raw === null) return null;
   const r = raw as Record<string, unknown>;
+  // Some models (e.g. Kimi K2) return numeric ids instead of string ids like "Idea 01".
+  // Coerce to string for compatibility.
+  if (typeof r.id === 'number') {
+    r.id = 'Idea ' + String(r.id).padStart(2, '0');
+  }
   if (typeof r.id !== 'string') return null;
   const note = typeof r.note === 'string' ? r.note : '';
   const s = r.scores;

src/core/budget/budget-tracker.ts

@@ -155,6 +155,20 @@ const FREE_LOCAL_EMBED_PROVIDERS: ReadonlySet<string> = new Set([
   'llama-server',
 ]);
 
+/**
+ * Provider id prefixes whose chat models run on local inference (electricity,
+ * not API tokens) and so price at $0. Without this, a `--max-cost`-bounded
+ * brainstorm/lsd configured for a local chat provider TX2 hard-fails because
+ * lookupPricing has no entry for them. Matched against the provider half
+ * of the `provider:model` string.
+ *
+ * Sibling to FREE_LOCAL_EMBED_PROVIDERS and FREE_LOCAL_RERANK_PROVIDERS.
+ */
+const FREE_LOCAL_CHAT_PROVIDERS: ReadonlySet<string> = new Set([
+  'ollama',
+  'llama-server',
+]);
+
 /**
  * Look up `modelId` in the chat or embedding pricing maps. Returns a
  * per-1M-token price tuple, or null when unknown.
@@ -201,6 +215,9 @@ function lookupPricing(modelId: string, kind: BudgetKind): ModelPricing | null {
   if (kind === 'rerank' && providerId && FREE_LOCAL_RERANK_PROVIDERS.has(providerId)) {
     return { input: 0, output: 0 };
   }
+  if (kind === 'chat' && providerId && FREE_LOCAL_CHAT_PROVIDERS.has(providerId)) {
+    return { input: 0, output: 0 };
+  }
   return null;
 }
 

src/core/think/index.ts

@@ -815,7 +815,7 @@ function buildGracefulMessage(modelStr: string): {
     type: 'message',
     role: 'assistant',
     model: modelStr,
-    content: [{ type: 'text', text: '(no LLM available — set anthropic_api_key via gbrain config or ANTHROPIC_API_KEY env)' }],
+    content: [{ type: 'text', text: '(no LLM available — set chat_model, expansion_model, or anthropic_api_key via gbrain config)' }],
     usage: { input_tokens: 0, output_tokens: 0 },
     stop_reason: 'end_turn',
   };