lynkr 9.4.5 → 9.5.0

package/README.md

@@ -545,6 +545,28 @@ TOOL_INJECTION_ENABLED=false
 CODE_MODE_ENABLED=true
 ```
 
+Always-on (no config): **smart tool selection** (server mode), **RTK tool-result
+compression** (test/git/grep/lint/build/JSON output), **MCP tool dedup** (drops
+built-in WebSearch/WebFetch when an Exa/Tavily MCP tool is present), and
+**request bypass** (Claude CLI Warmup / title-extraction calls are answered
+locally, never hitting a provider).
+
+Optional **terse-output mode** to cut *output* tokens:
+```bash
+CAVEMAN_ENABLED=true        # off by default — nudges the model to be concise
+CAVEMAN_LEVEL=lite          # lite | full | ultra
+```
+
+### Cost tracking & model pricing
+Per-request cost is computed from a model-pricing registry (LiteLLM → models.dev,
+cached 24h) and recorded in telemetry. Models the registry doesn't know record
+`cost_usd=null` (logged once) rather than a fabricated price. Pin prices for
+unknown models:
+```bash
+# Per-1M-token USD prices, JSON keyed by model name
+MODEL_PRICE_OVERRIDES={"my-model":{"input":0.5,"output":1.5}}
+```
+
 ### Memory System (Titans-inspired)
 ```bash
 MEMORY_ENABLED=true
@@ -652,35 +674,45 @@ npm start
 
 ## Benchmark Results
 
-Measured on real agentic coding workloads (Claude Code / Cursor sessions) with Ollama, Moonshot, and Azure OpenAI backends. Run with `node benchmark-tier-routing.js`.
+Head-to-head against **LiteLLM** on the **same backends** (Ollama `minimax-m2.5`, Moonshot, Azure OpenAI), 9 scenarios across 4 feature categories. Apples-to-apples comparison is Lynkr vs LiteLLM **billed tokens on the same scenario**. Run with `node benchmark-tier-routing.js`.
 
-### Token compression
+> _Run: June 5, 2026 · Lynkr v9.3.2 · LiteLLM v1.87.1 · macOS, Apple Silicon._
 
-| Scenario | Tokens without Lynkr | Tokens with Lynkr | Reduction |
+### Token reduction (vs LiteLLM, same model & prompt)
+
+| Mechanism | Lynkr | LiteLLM | Result |
 |---|---|---|---|
-| 14-tool request (read task) | 1,042 | **547** | **47%** |
-| 14-tool request (write task) | 1,043 | **412** | **60%** |
-| Large JSON grep result (60 items) | 3,458 | **427** | **87.6%** |
+| Smart tool selection (14 tools) | **959** tokens · $0.0044 | 2,085 tokens · $0.0091 | **53% fewer tokens, 52% cheaper** |
+| TOON compression (60-item grep JSON) | **427** tokens · $0.009 | 3,458 tokens · $0.018 | **87.6% fewer tokens, 50% cheaper** |
 
-Lynkr strips irrelevant tool schemas before forwarding (smart tool selection) and binary-compresses large JSON tool results (TOON) — both happen in-process with no added latency.
+Lynkr strips irrelevant tool schemas (smart tool selection) and binary-compresses large JSON tool results (TOON) — both in-process, no added latency.
 
 ### Semantic cache
 
 | | Tokens billed | Response time |
 |---|---|---|
 | First call (cold) | 2,857 | 1,891ms |
-| **Second call — paraphrased, cache hit** | **0** | **171ms** |
+| **Second call — paraphrased, cache hit** | **0** (served from cache) | **171ms (11× faster)** |
 
-Near-identical prompts return cached responses in 171ms. Zero tokens billed on a cache hit.
+Near-identical prompts return cached responses in 171ms. Zero model tokens billed on a cache hit.
 
 ### Tier routing
 
-| Request | Routed to |
-|---|---|
-| "What does git stash do?" | SIMPLE → local model (free) |
-| JWT vs cookies security analysis | COMPLEX → cloud model (correct) |
+| Request | Lynkr routes to | LiteLLM routes to |
+|---|---|---|
+| "What does git stash do?" | `minimax-m2.5` (local, free) | Ollama (local) |
+| JWT vs cookies security analysis | `moonshot` (cloud — correct) | **Ollama (local — wrong call)** |
+
+Lynkr scores each request on 15 dimensions (token count, code complexity, reasoning markers, risk signals, agentic patterns) and escalates automatically. LiteLLM's `cost-based-routing` sends everything to the cheapest model regardless of complexity.
+
+### Cost projection (100,000 requests/month, same backend)
+
+| | Monthly cost | vs LiteLLM |
+|---|---|---|
+| LiteLLM | ~$818 | baseline |
+| **Lynkr** | **~$409** | **~50% cheaper** |
 
-Lynkr scores each request on 15 dimensions (token count, code complexity, reasoning markers, risk signals, agentic patterns) and routes automatically. No caller changes needed.
+_Based on a tool-heavy agentic session (TOON scenario). On equal footing — same provider, same model — Lynkr is cheaper due to token optimization._
 
 → [Full benchmark report with methodology](BENCHMARK_REPORT.md)
 

package/install.sh

@@ -108,8 +108,24 @@ clone_or_update() {
 install_dependencies() {
     print_info "Installing dependencies..."
     cd "$INSTALL_DIR"
-    npm install --production
+    # --omit=dev keeps optionalDependencies (better-sqlite3, hnswlib-node,
+    # tree-sitter) which back telemetry, the memory store and routing ML.
+    # The postinstall hook (scripts/check-native.js) verifies the native ABI
+    # and rebuilds if Node was upgraded — best-effort, never fails the install.
+    npm install --omit=dev
     print_success "Dependencies installed"
+
+    # Native optional modules need a C/C++ toolchain only if no prebuilt binary
+    # is available for this platform. They degrade gracefully if absent.
+    if ! node -e "const D=require('better-sqlite3'); new D(':memory:').close()" >/dev/null 2>&1; then
+        print_warning "Native module 'better-sqlite3' is not loadable."
+        echo "     Telemetry, the memory store and sessions need it. To enable:"
+        echo "       - Ensure a build toolchain is present (Xcode CLT on macOS, build-essential + python3 on Linux), then:"
+        echo "       - ${BLUE}cd $INSTALL_DIR && npm run rebuild-native${NC}"
+        echo "     Lynkr still runs without it (those features stay disabled)."
+    else
+        print_success "Native modules OK (telemetry, memory, sessions enabled)"
+    fi
 }
 
 # Create default .env file
@@ -131,7 +147,7 @@ create_env_file() {
 MODEL_PROVIDER=ollama
 
 # Server Configuration
-PORT=8080
+PORT=8081
 
 # Ollama Configuration (default for local development)
 OLLAMA_MODEL=qwen2.5-coder:7b
@@ -161,7 +177,7 @@ EOF
         print_info "📝 Configuration ready! Key settings:"
         echo "     • Default provider: Ollama (local, offline)"
         echo "     • Memory system: Enabled (learns from conversations)"
-        echo "     • Port: 8080"
+        echo "     • Port: 8081"
         echo ""
         print_warning "To use cloud providers (Databricks/OpenAI/Azure):"
         echo "     Edit: ${BLUE}nano $INSTALL_DIR/.env${NC}"
@@ -220,7 +236,7 @@ print_next_steps() {
     echo "     ${BLUE}lynkr${NC}"
     echo ""
     echo "  3. Configure Claude Code CLI:"
-    echo "     ${BLUE}export ANTHROPIC_BASE_URL=http://localhost:8080${NC}"
+    echo "     ${BLUE}export ANTHROPIC_BASE_URL=http://localhost:8081${NC}"
     echo "     ${BLUE}claude${NC}"
     echo ""
     echo "  ${YELLOW}Option B: Use Cloud Providers (Databricks/OpenAI/Azure)${NC}"
@@ -238,7 +254,7 @@ print_next_steps() {
     echo "     ${BLUE}lynkr${NC}"
     echo ""
     echo "  3. Configure Claude Code CLI:"
-    echo "     ${BLUE}export ANTHROPIC_BASE_URL=http://localhost:8080${NC}"
+    echo "     ${BLUE}export ANTHROPIC_BASE_URL=http://localhost:8081${NC}"
     echo "     ${BLUE}export ANTHROPIC_API_KEY=any-non-empty-value${NC}  ${GREEN}← Placeholder${NC}"
     echo "     ${BLUE}claude${NC}"
     echo ""

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "lynkr",
-  "version": "9.4.5",
+  "version": "9.5.0",
   "description": "Self-hosted LLM gateway and tier-routing proxy for Claude Code, Cursor, and Codex. Routes across Ollama, AWS Bedrock, OpenRouter, Databricks, Azure OpenAI, llama.cpp, and LM Studio with prompt caching, MCP tools, and 60-80% cost savings.",
   "main": "index.js",
   "bin": {
@@ -8,13 +8,15 @@
     "lynkr-setup": "scripts/setup.js"
   },
   "scripts": {
+    "postinstall": "node scripts/check-native.js",
+    "rebuild-native": "node scripts/check-native.js",
     "prestart": "node -e \"if(process.env.HEADROOM_ENABLED==='true'&&process.env.HEADROOM_DOCKER_ENABLED!=='false'){process.exit(0)}else{process.exit(1)}\" && docker compose --profile headroom up -d --build headroom 2>/dev/null || echo 'Headroom skipped (disabled or Docker not running)'",
     "start": "node index.js 2>&1 | npx pino-pretty --sync",
     "stop": "node -e \"if(process.env.HEADROOM_ENABLED==='true'&&process.env.HEADROOM_DOCKER_ENABLED!=='false'){process.exit(0)}else{process.exit(1)}\" && docker compose --profile headroom down || echo 'Headroom skipped (disabled or Docker not running)'",
     "dev": "nodemon index.js",
     "lint": "eslint src index.js",
     "test": "npm run test:unit && npm run test:performance",
-    "test:unit": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/routing.test.js test/hybrid-routing-integration.test.js test/web-tools.test.js test/passthrough-mode.test.js test/openrouter-error-resilience.test.js test/format-conversion.test.js test/azure-openai-config.test.js test/azure-openai-format-conversion.test.js test/azure-openai-routing.test.js test/azure-openai-streaming.test.js test/azure-openai-error-resilience.test.js test/azure-openai-integration.test.js test/openai-integration.test.js test/toon-compression.test.js test/llamacpp-integration.test.js test/resilience.test.js test/telemetry-routing.test.js test/memory/store.test.js test/memory/surprise.test.js test/memory/extractor.test.js test/memory/search.test.js test/memory/retriever.test.js test/distill.test.js test/large-payload.test.js test/code-mode.test.js test/prompt-cache-injection.test.js test/risk-analyzer.test.js test/interaction-block.test.js test/preflight.test.js",
+    "test:unit": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/routing.test.js test/hybrid-routing-integration.test.js test/web-tools.test.js test/passthrough-mode.test.js test/openrouter-error-resilience.test.js test/format-conversion.test.js test/azure-openai-config.test.js test/azure-openai-format-conversion.test.js test/azure-openai-routing.test.js test/azure-openai-streaming.test.js test/azure-openai-error-resilience.test.js test/azure-openai-integration.test.js test/openai-integration.test.js test/toon-compression.test.js test/llamacpp-integration.test.js test/resilience.test.js test/telemetry-routing.test.js test/memory/store.test.js test/memory/surprise.test.js test/memory/extractor.test.js test/memory/search.test.js test/memory/retriever.test.js test/distill.test.js test/large-payload.test.js test/code-mode.test.js test/prompt-cache-injection.test.js test/risk-analyzer.test.js test/interaction-block.test.js test/preflight.test.js test/token-reduction.test.js test/session-affinity.test.js test/model-registry-cost.test.js",
     "test:memory": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/memory/store.test.js test/memory/surprise.test.js test/memory/extractor.test.js test/memory/search.test.js test/memory/retriever.test.js",
     "test:new-features": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/passthrough-mode.test.js test/openrouter-error-resilience.test.js test/format-conversion.test.js",
     "test:performance": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/hybrid-routing-performance.test.js && DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/performance-tests.js",

package/public/dashboard.html

@@ -244,6 +244,7 @@ const App = {
     const t = d.today;
     const s = d.stats;
 
+    const tierLabel = t => t === 'default' ? 'default' : String(t).toLowerCase();
     const providerCards = d.providers.length === 0
       ? `<p class="text-slate-500 text-sm">No providers configured</p>`
       : d.providers.map(p => `
@@ -251,10 +252,21 @@ const App = {
           <div class="flex items-center gap-2">
             <span class="status-dot ${providerDot(p.type)}"></span>
             <span class="text-sm font-medium text-slate-200">${p.name}</span>
+            ${(p.tiers || []).map(t => `<span class="badge bg-slate-600/60 text-slate-300">${tierLabel(t)}</span>`).join('')}
           </div>
           <span class="text-xs ${p.type === 'local' ? 'text-green-400' : 'text-blue-400'}">${p.type}</span>
         </div>`).join('');
 
+    const providerWarnings = (d.providerWarnings || []).map(w => `
+        <div class="flex items-center justify-between bg-amber-500/10 border border-amber-500/30 rounded-lg px-4 py-3">
+          <div class="flex items-center gap-2">
+            <span class="text-amber-400 text-sm">⚠</span>
+            <span class="text-sm font-medium text-amber-200">${w.name}</span>
+            ${(w.tiers || []).map(t => `<span class="badge bg-amber-500/20 text-amber-300">${tierLabel(t)}</span>`).join('')}
+          </div>
+          <span class="text-xs text-amber-400">no credentials</span>
+        </div>`).join('');
+
     const recentRows = (d.recentRequests || []).map(r => `
       <tr class="table-row border-b border-slate-700/50">
         <td class="py-2 px-3 text-xs text-slate-500">${fmt.ago(r.timestamp)}</td>
@@ -279,7 +291,7 @@ const App = {
         <!-- Providers -->
         ${card(`
           <h3 class="text-sm font-semibold text-slate-300 mb-3">Configured Providers</h3>
-          <div class="flex flex-col gap-2">${providerCards}</div>
+          <div class="flex flex-col gap-2">${providerCards}${providerWarnings}</div>
         `)}
 
         <!-- 24h Stats -->

package/scripts/check-native.js

@@ -0,0 +1,97 @@
+#!/usr/bin/env node
+/**
+ * Native module ABI guard (postinstall).
+ *
+ * better-sqlite3 (and the other native optionalDependencies) are compiled
+ * against a specific Node ABI. When Node is upgraded, the prebuilt/compiled
+ * binary stops loading with:
+ *
+ *   "was compiled against a different Node.js version using
+ *    NODE_MODULE_VERSION 115. This version of Node.js requires
+ *    NODE_MODULE_VERSION 141."
+ *
+ * The failure is silent at runtime — telemetry, request logs, and the memory
+ * store all sit behind try/catch and simply go empty. This probe detects the
+ * mismatch and rebuilds the native modules so it self-heals on `npm install`.
+ *
+ * It is intentionally best-effort: it NEVER exits non-zero, so it can't break
+ * `npm install` on machines without a build toolchain (the modules are
+ * optional and the app degrades gracefully without them).
+ */
+
+const { execSync } = require("child_process");
+
+// Native optionalDependencies that are ABI-sensitive. If Node changed, all of
+// them are stale, so we rebuild the set in one pass.
+const NATIVE_DEPS = [
+  "better-sqlite3",
+  "hnswlib-node",
+  "tree-sitter",
+  "tree-sitter-javascript",
+  "tree-sitter-python",
+  "tree-sitter-typescript",
+];
+
+function log(msg) {
+  console.log(`[check-native] ${msg}`);
+}
+
+/**
+ * Probe better-sqlite3 — the canary. `require()` alone is not enough: the
+ * native addon only loads when a Database is instantiated.
+ * @returns {"ok"|"absent"|"mismatch"}
+ */
+function probe() {
+  let Database;
+  try {
+    Database = require("better-sqlite3");
+  } catch (err) {
+    if (err && err.code === "MODULE_NOT_FOUND") return "absent";
+    return "mismatch";
+  }
+  try {
+    const db = new Database(":memory:");
+    db.close();
+    return "ok";
+  } catch (err) {
+    if (/NODE_MODULE_VERSION|different Node\.js version|invalid ELF|dlopen|\.node/i.test(err.message || "")) {
+      return "mismatch";
+    }
+    // Some other instantiation error — not an ABI issue we can fix by rebuild.
+    return "ok";
+  }
+}
+
+function main() {
+  const status = probe();
+
+  if (status === "absent") {
+    // Optional dependency not installed (e.g. build skipped). Nothing to do.
+    return;
+  }
+  if (status === "ok") {
+    return;
+  }
+
+  log("native module ABI mismatch detected (Node was likely upgraded). Rebuilding native modules…");
+  try {
+    execSync(`npm rebuild ${NATIVE_DEPS.join(" ")}`, { stdio: "inherit" });
+  } catch {
+    log("rebuild did not complete (a build toolchain may be missing). Continuing — native features will be disabled until you run: npm rebuild better-sqlite3");
+    return;
+  }
+
+  // Re-probe to report the outcome.
+  if (probe() === "ok") {
+    log("native modules rebuilt successfully.");
+  } else {
+    log("native modules still not loadable after rebuild. Run `npm rebuild better-sqlite3` manually.");
+  }
+}
+
+try {
+  main();
+} catch (err) {
+  // Never fail the install.
+  log(`skipped (${err.message})`);
+}

package/src/clients/databricks.js

@@ -1506,10 +1506,16 @@ async function invokeMoonshot(body) {
     "claude-haiku-4-5-20251001": "kimi-k2-turbo-preview",
     "claude-haiku-4-5": "kimi-k2-turbo-preview",
     "claude-3-haiku": "kimi-k2-turbo-preview",
+    // moonshot-v1-auto 400s with "tokenization failed" (its server-side auto
+    // context-size pass fails on large tool-bearing payloads). Remap to a
+    // fixed model that's broadly available on api.moonshot.ai.
+    "moonshot-v1-auto": "moonshot-v1-128k",
   };
 
   const requestedModel = body._tierModel || body.model || config.moonshot.model;
-  const mappedModel = modelMap[requestedModel] || config.moonshot.model || "kimi-k2-turbo-preview";
+  let mappedModel = modelMap[requestedModel] || config.moonshot.model || "kimi-k2-turbo-preview";
+  // Guard against the deprecated auto model arriving via config too.
+  if (mappedModel === "moonshot-v1-auto") mappedModel = "moonshot-v1-128k";
 
   // Convert messages using existing utility
   const messages = convertAnthropicMessagesToOpenRouter(body.messages || []);
@@ -1522,12 +1528,18 @@ async function invokeMoonshot(body) {
     messages.unshift({ role: "system", content: systemContent });
   }
 
+  // kimi-k2.x (k2.5 / k2.6 …) are thinking models that only accept
+  // temperature: 1 — any other value 400s with "invalid temperature".
+  const isKimiThinking = /^kimi-k2/i.test(mappedModel);
+
   const moonshotBody = {
     model: mappedModel,
     messages,
     max_tokens: body.max_tokens || 16384,
-    temperature: body.temperature ?? 0.7,
-    top_p: body.top_p ?? 1.0,
+    // kimi-k2.x thinking models pin sampling params: temperature must be 1
+    // and top_p must be 0.95 — any other value 400s.
+    temperature: isKimiThinking ? 1 : (body.temperature ?? 0.7),
+    top_p: isKimiThinking ? 0.95 : (body.top_p ?? 1.0),
     stream: false,  // Force non-streaming - OpenAI SSE to Anthropic SSE conversion not implemented
   };
 
@@ -2027,6 +2039,65 @@ async function invokeCodex(body) {
   };
 }
 
+/**
+ * Compute request cost in USD from model pricing × token usage.
+ * Registry returns per-1M-token prices ({ input, output }); returns null when
+ * pricing is unknown so we don't record misleading zeros.
+ */
+const _unknownCostWarned = new Set();
+function computeCostUsd(model, inputTokens, outputTokens) {
+  try {
+    const { getModelRegistrySync } = require("../routing/model-registry");
+    const reg = getModelRegistrySync && getModelRegistrySync();
+    const cost = reg?.getCost?.(model);
+    if (!cost) return null;
+    // Unknown model → record null (not a fabricated default), warn once so the
+    // gap is visible and can be fixed via MODEL_PRICE_OVERRIDES.
+    if (cost.unknown) {
+      if (model && !_unknownCostWarned.has(model)) {
+        _unknownCostWarned.add(model);
+        logger.warn({ model }, "[Cost] No pricing for model — recording cost_usd=null. Set MODEL_PRICE_OVERRIDES to fix.");
+      }
+      return null;
+    }
+    if (cost.input == null && cost.output == null) return null;
+    const inUsd = ((inputTokens || 0) / 1e6) * (cost.input || 0);
+    const outUsd = ((outputTokens || 0) / 1e6) * (cost.output || 0);
+    return Number((inUsd + outUsd).toFixed(6));
+  } catch {
+    return null;
+  }
+}
+
+// Telemetry prompt/response text is always captured (truncated) to build the
+// routing ML training corpus. Stored locally in .lynkr/telemetry.db only.
+const TELEMETRY_TEXT_MAXLEN = 2000;
+
+/** Flatten the latest user message to plain text (for telemetry capture). */
+function captureRequestText(body) {
+  const messages = body?.messages;
+  if (!Array.isArray(messages)) return null;
+  for (let i = messages.length - 1; i >= 0; i--) {
+    const m = messages[i];
+    if (m?.role !== "user") continue;
+    let text = "";
+    if (typeof m.content === "string") text = m.content;
+    else if (Array.isArray(m.content)) {
+      text = m.content.filter((b) => b?.type === "text").map((b) => b.text || "").join(" ");
+    }
+    if (text) return text.slice(0, TELEMETRY_TEXT_MAXLEN);
+  }
+  return null;
+}
+
+/** Flatten an Anthropic response's text blocks to plain text (for telemetry). */
+function captureResponseText(resultJson) {
+  const content = resultJson?.content;
+  if (!Array.isArray(content)) return null;
+  const text = content.filter((b) => b?.type === "text").map((b) => b.text || "").join(" ");
+  return text ? text.slice(0, TELEMETRY_TEXT_MAXLEN) : null;
+}
+
 async function invokeModel(body, options = {}) {
   const { determineProviderSmart, isFallbackEnabled, getFallbackProvider } = require("./routing");
   const metricsCollector = getMetricsCollector();
@@ -2233,6 +2304,9 @@ async function invokeModel(body, options = {}) {
       circuit_breaker_state: breaker.state,
       quality_score: qualityScore,
       tokens_per_second: outputTokens && latency > 0 ? outputTokens / (latency / 1000) : null,
+      cost_usd: computeCostUsd(routingDecision.model || body._tierModel, inputTokens, outputTokens),
+      request_text: captureRequestText(body),
+      response_text: captureResponseText(result.json),
     });
 
     // Return result with provider info and routing decision for headers
@@ -2394,6 +2468,9 @@ async function invokeModel(body, options = {}) {
           { status_code: 200, output_tokens: fbOutputTokens, tool_calls_made: fbToolCalls, was_fallback: true, retry_count: 0, latency_ms: Date.now() - startTime }
         ),
         tokens_per_second: fbOutputTokens && fallbackLatency > 0 ? fbOutputTokens / (fallbackLatency / 1000) : null,
+        cost_usd: computeCostUsd(routingDecision.model || body._tierModel, fbInputTokens, fbOutputTokens),
+        request_text: captureRequestText(body),
+        response_text: captureResponseText(fallbackResult.json),
       });
 
       // Return result with actual provider used (fallback provider) and routing decision

package/src/clients/openrouter-utils.js

@@ -176,6 +176,21 @@ function convertAnthropicMessagesToOpenRouter(anthropicMessages) {
     }
   }
 
+  // Kimi/Moonshot (and some OpenAI-compatible APIs) reject a message whose
+  // content is an empty string with "Invalid request: tokenization failed".
+  // This happens when a turn had only non-text blocks (thinking / image /
+  // stripped content) and flattened to "". Replace empty/whitespace-only
+  // content with a single space — but never touch an assistant message that
+  // carries tool_calls, where content: null is intentional and required.
+  for (const m of converted) {
+    if (m.role === 'tool') continue;
+    const hasToolCalls = Array.isArray(m.tool_calls) && m.tool_calls.length > 0;
+    if (hasToolCalls) continue;
+    if (typeof m.content !== 'string' || m.content.trim() === '') {
+      m.content = ' ';
+    }
+  }
+
   // Log the converted messages for debugging
   logger.debug({
     inputCount: anthropicMessages.length,

package/src/config/index.js

@@ -208,6 +208,11 @@ const tokenBudgetWarning = Number.parseInt(process.env.TOKEN_BUDGET_WARNING ?? "
 const tokenBudgetMax = Number.parseInt(process.env.TOKEN_BUDGET_MAX ?? "180000", 10);
 const tokenBudgetEnforcement = process.env.TOKEN_BUDGET_ENFORCEMENT !== "false"; // default true
 
+// Caveman terse-output injection (opt-in, off by default)
+const cavemanEnabled = process.env.CAVEMAN_ENABLED === "true";
+const cavemanLevel = (process.env.CAVEMAN_LEVEL ?? "lite").toLowerCase();
+
+
 // TOON payload compression (opt-in)
 const toonEnabled = process.env.TOON_ENABLED === "true"; // default false
 const toonMinBytes = Number.parseInt(process.env.TOON_MIN_BYTES ?? "4096", 10);
@@ -641,6 +646,10 @@ var config = {
   toolResultCompression: {
     enabled: true,
   },
+  caveman: {
+    enabled: cavemanEnabled,
+    level: cavemanLevel,
+  },
   server: {
     jsonLimit: process.env.REQUEST_JSON_LIMIT ?? "1gb",
   },

package/src/context/caveman.js

@@ -0,0 +1,94 @@
+/**
+ * Caveman Terse-Output Injector
+ *
+ * Appends a brevity instruction to the system prompt so the model produces
+ * terser responses, reducing OUTPUT tokens. Opt-in and off by default — it
+ * changes model behavior, so it's only applied when explicitly enabled.
+ *
+ * Enable with CAVEMAN_ENABLED=true. Level via CAVEMAN_LEVEL=lite|full|ultra
+ * (default: lite). Adapted from 9router's caveman injector / the caveman skill
+ * (https://github.com/JuliusBrussee/caveman).
+ *
+ * @module context/caveman
+ */
+
+const config = require("../config");
+const logger = require("../logger");
+
+const LEVELS = ["lite", "full", "ultra"];
+
+// Shared guardrails so brevity never corrupts the substance that matters.
+const BOUNDARIES =
+  "Code blocks, file paths, commands, errors, URLs: keep exact. " +
+  "Security warnings, irreversible-action confirmations, and multi-step ordered " +
+  "sequences: write in full normal prose. Resume terse style afterward.";
+
+const EXAMPLES =
+  'Not: "Sure! I\'d be happy to help. The issue is likely caused by..." ' +
+  'Yes: "Bug in auth middleware. Token expiry uses `<` not `<=`. Fix:"';
+
+const PERSISTENCE = "Apply this to every response unless a guardrail above applies.";
+
+const PROMPTS = {
+  lite: [
+    "Respond tersely. Keep grammar and full sentences but drop filler, hedging, and pleasantries (just/really/basically/sure/of course/I'd be happy to).",
+    "Pattern: state the thing, the action, the reason. Then the next step.",
+    EXAMPLES,
+    BOUNDARIES,
+    PERSISTENCE,
+  ].join(" "),
+
+  full: [
+    "Respond like a terse caveman. All technical substance stays exact; only fluff dies.",
+    "Drop articles (a/an/the), filler (just/really/basically/actually/simply), pleasantries, and hedging. Fragments OK. Prefer short synonyms (big not extensive, fix not implement a solution for).",
+    "Pattern: [thing] [action] [reason]. [next step].",
+    EXAMPLES,
+    BOUNDARIES,
+    PERSISTENCE,
+  ].join(" "),
+
+  ultra: [
+    "Respond ultra-terse. Maximum compression. Telegraphic.",
+    "Abbreviate (DB/auth/config/req/res/fn/impl), strip conjunctions, use arrows for causality (X → Y). One word when one word is enough.",
+    "Pattern: [thing] → [result]. [fix].",
+    EXAMPLES,
+    BOUNDARIES,
+    PERSISTENCE,
+  ].join(" "),
+};
+
+const MARKER = "[brevity]";
+
+/** Resolve the configured level, falling back to "lite". */
+function resolveLevel(level) {
+  const l = String(level || config.caveman?.level || "lite").toLowerCase();
+  return LEVELS.includes(l) ? l : "lite";
+}
+
+/**
+ * Append the brevity instruction to a system prompt string.
+ * Idempotent — won't double-inject if the marker is already present.
+ *
+ * @param {string} system - Existing system prompt (may be empty).
+ * @param {object} [opts]
+ * @param {boolean} [opts.enabled] - Override config enablement.
+ * @param {string} [opts.level] - Override level.
+ * @returns {string} system prompt, possibly with brevity instruction appended.
+ */
+function injectCaveman(system, opts = {}) {
+  const enabled = opts.enabled ?? config.caveman?.enabled === true;
+  if (!enabled) return system || "";
+
+  const base = system || "";
+  if (base.includes(MARKER)) return base;
+
+  const level = resolveLevel(opts.level);
+  const instruction = `\n\n${MARKER} ${PROMPTS[level]}`;
+  logger.debug({ level }, "[Caveman] Injected brevity instruction into system prompt");
+  return base + instruction;
+}
+
+module.exports = {
+  injectCaveman,
+  LEVELS,
+};