fix: correct -llama to -lllama, add missing paths, fix CI, bump golangci

.github/workflows/ci.yml

@@ -16,16 +16,22 @@ jobs:
           go-version: "1.23"
 
       - name: Build
-        run: go build -o /dev/null ./cmd/coginfer
+        run: CGO_ENABLED=0 go build -o /dev/null ./cmd/coginfer
 
       - name: Lint
-        uses: golangci/golangci-lint-action@v6
+        uses: golangci/golangci-lint-action@v9
+        env:
+          CGO_ENABLED: 0
         with:
           version: latest
           args: --timeout=3m
 
       - name: Test
+        env:
+          CGO_ENABLED: 0
         run: go test ./... -v -count=1
 
       - name: Vet
+        env:
+          CGO_ENABLED: 0
         run: go vet ./...

AGENTS.md

@@ -20,7 +20,7 @@ internal/llm/            — Backend interface + implementations
   ├── cgobackend.go      — CgoBackend: Backend impl calling bridge functions (build tag: cgo)
   └── loadopts.go        — LoadOptions{NumCtx, GPULayers, Threads}
 internal/model/          — Model scanning and metadata (.gguf file discovery)
-vendor/llama.cpp/        — Git submodule, pinned to b9842
+vendor/llama.cpp/        — Manually cloned (NOT a git submodule — no `.gitmodules` file)
 ```
 
 ### coginfer (Wide Model)
@@ -30,18 +30,27 @@ LLM inference server linking llama.cpp via a vendored CGo bridge, exposing an Ol
 #### Build
 
 ```bash
-# With CGo (production — requires cmake + gcc + submodule)
+# With CGo (production — requires cmake + gcc + llama.cpp cloned into vendor/)
 cd vendor/llama.cpp && cmake -B build -DLLAMA_NO_ACCELERATE=1 \
   -DLLAMA_STATIC=1 -DLLAMA_NATIVE=0 \
   -DBUILD_SHARED_LIBS=0 -DLLAMA_BUILD_TESTS=0 \
-  -DLLAMA_BUILD_EXAMPLES=0 -DLLAMA_BUILD_SERVER=0
-cmake --build build --config Release -j$(nproc)
+  -DLLAMA_BUILD_EXAMPLES=0 -DLLAMA_BUILD_SERVER=0 \
+  -DCMAKE_ARCHIVE_OUTPUT_DIRECTORY="$PWD/build"
+cmake --build build --target llama --config Release -j$(nproc)
 cd ../.. && CGO_ENABLED=1 go build -tags=cgo -o bin/coginfer ./cmd/coginfer
 
 # Without CGo (CI, development — mock backend only)
 CGO_ENABLED=0 go build -o bin/coginfer ./cmd/coginfer
 ```
 
+**CGo flags** (in `internal/llm/bridge.go`):
+- `-lllama` — NOT `-llama` (cmake target `llama` → `libllama.a`)
+- `-L.../build/src` — modern llama.cpp puts static archives in `build/src/`
+  (workaround: `CMAKE_ARCHIVE_OUTPUT_DIRECTORY` puts it in `build/`)
+- `-I.../ggml/include` — `llama.h` includes `ggml.h`, `ggml-cpu.h`, etc.
+- `CGO_LDFLAGS` is appended **before** `#cgo LDFLAGS` by Go toolchain;
+  `-L` in env var establishes search path before `-l` in source
+
 #### Usage
 
 ```bash
@@ -51,9 +60,6 @@ CGO_ENABLED=0 go build -o bin/coginfer ./cmd/coginfer
 # Start with CGo llama.cpp bridge (production)
 ./bin/coginfer --backend cgo --models /cognitiveos/models
 
-# Start with CGo llama.cpp bridge (production)
-./bin/coginfer --backend cgo --models /cognitiveos/models
-
 # Custom port and log file
 ./bin/coginfer --addr 127.0.0.1:11434 --log /cognitiveos/logs/inference.log
 ```
@@ -77,13 +83,12 @@ CGO_ENABLED=0 go build -o bin/coginfer ./cmd/coginfer
 
 - **mock**: Simulated token generation with delays, no external dependencies. Default for development. Always available.
 - **cgo**: In-process llama.cpp via CGo bridge. Requires `CGO_ENABLED=1` and `vendor/llama.cpp/build/libllama.a`. Production default.
-- **cgo**: In-process llama.cpp via CGo bridge. Requires `CGO_ENABLED=1` and `vendor/llama.cpp/build/libllama.a`. Production default.
 
 #### Build Tags
 
 - `bridge.go`, `cgobackend.go`, `backend_cgo.go` — `//go:build cgo`
 - `backend_stub.go` — `//go:build !cgo`
-- CI runs `CGO_ENABLED=0`, excludes all CGo files automatically
+- CI runs `CGO_ENABLED=0` (Ubuntu defaults to `CGO_ENABLED=1`, which includes `cgo` tag — must be explicit)
 - `MockBackend` has no build tag — always available
 
 ### cograw (Raw Model — Firmware Guardrail)

cmd/coginfer/main.go

@@ -20,7 +20,7 @@ func main() {
 		if err != nil {
 			log.Fatal(err)
 		}
-		defer f.Close()
+		defer func() { _ = f.Close() }()
 		log.SetOutput(f)
 	}
 

cmd/cograw/main.go

@@ -181,7 +181,7 @@ var rawLog *log.Logger
 
 func initRawLog(path string) {
 	dir := filepath.Dir(path)
-	os.MkdirAll(dir, 0755)
+	_ = os.MkdirAll(dir, 0755)
 	f, err := os.OpenFile(path, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0640)
 	if err != nil {
 		log.Printf("WARN: cannot open raw audit log %s: %v", path, err)
@@ -215,7 +215,7 @@ func main() {
 		if err != nil {
 			log.Fatal(err)
 		}
-		defer f.Close()
+		defer func() { _ = f.Close() }()
 		log.SetOutput(f)
 	}
 
@@ -227,7 +227,7 @@ func main() {
 	}
 	logAudit("startup", fmt.Sprintf("raw model loaded: %s", *modelPath))
 
-	os.Remove(*socketPath)
+	_ = os.Remove(*socketPath)
 	addr, err := net.ResolveUnixAddr("unix", *socketPath)
 	if err != nil {
 		log.Fatalf("resolve addr: %v", err)
@@ -240,7 +240,7 @@ func main() {
 	if err := os.Chmod(*socketPath, 0600); err != nil {
 		log.Fatalf("chmod: %v", err)
 	}
-	defer os.Remove(*socketPath)
+	defer func() { _ = os.Remove(*socketPath) }()
 
 	log.Printf("cograw ready on %s (model: %s)", *socketPath, rm.model)
 
@@ -250,7 +250,7 @@ func main() {
 	go func() {
 		<-sigCh
 		log.Println("shutting down")
-		listener.Close()
+		_ = listener.Close()
 	}()
 
 	for {
@@ -283,7 +283,7 @@ func (r *RawModel) verifyModel(path string) error {
 }
 
 func handleConn(conn *net.UnixConn, rm *RawModel) {
-	defer conn.Close()
+	defer func() { _ = conn.Close() }()
 
 	decoder := json.NewDecoder(conn)
 	encoder := json.NewEncoder(conn)
@@ -295,7 +295,7 @@ func handleConn(conn *net.UnixConn, rm *RawModel) {
 		}
 
 		resp := dispatch(call, rm)
-		encoder.Encode(resp)
+		_ = encoder.Encode(resp)
 	}
 }
 
@@ -499,12 +499,12 @@ func readMemAvailableMB() (int64, int64) {
 	for _, line := range lines {
 		if strings.HasPrefix(line, "MemTotal:") {
 			var kb int64
-			fmt.Sscanf(line, "MemTotal: %d kB", &kb)
+			_, _ = fmt.Sscanf(line, "MemTotal: %d kB", &kb)
 			totalMB = kb / 1024
 		}
 		if strings.HasPrefix(line, "MemAvailable:") {
 			var kb int64
-			fmt.Sscanf(line, "MemAvailable: %d kB", &kb)
+			_, _ = fmt.Sscanf(line, "MemAvailable: %d kB", &kb)
 			availableMB = kb / 1024
 		}
 	}
@@ -519,10 +519,7 @@ func handleAudit(call RPCCall, rm *RawModel) RPCResp {
 
 	totalMB, freeMB := readMemAvailableMB()
 
-	allowed := true
-	if params.RequestedMB > 0 && params.RequestedMB > freeMB {
-		allowed = false
-	}
+	allowed := params.RequestedMB <= 0 || params.RequestedMB <= freeMB
 
 	return RPCResp{
 		JSONRPC: "2.0",

internal/llm/bridge.go

@@ -3,8 +3,8 @@
 package llm
 
 /*
-#cgo LDFLAGS: -L${SRCDIR}/../../vendor/llama.cpp/build -llama -lm -lstdc++
-#cgo CFLAGS: -I${SRCDIR}/../../vendor/llama.cpp/include
+#cgo LDFLAGS: -L${SRCDIR}/../../vendor/llama.cpp/build -L${SRCDIR}/../../vendor/llama.cpp/build/src -lllama -lm -lstdc++
+#cgo CFLAGS: -I${SRCDIR}/../../vendor/llama.cpp/include -I${SRCDIR}/../../vendor/llama.cpp/ggml/include
 
 #include <stdlib.h>
 #include "llama.h"

internal/llm/cgobackend.go

@@ -63,8 +63,8 @@ func (c *CgoBackend) Load(modelPath string, opts *LoadOptions) (*ModelInfo, erro
 		return nil, fmt.Errorf("E_MODEL_NOT_FOUND: %s: %w", modelPath, err)
 	}
 	magic := make([]byte, 4)
-	f.Read(magic)
-	f.Close()
+	_, _ = f.Read(magic)
+	_ = f.Close()
 	if string(magic) != "GGUF" {
 		return nil, fmt.Errorf("E_MODEL_LOAD_FAILED: invalid GGUF magic bytes in %s", modelPath)
 	}

internal/llm/llm_test.go

@@ -27,7 +27,7 @@ func TestMockBackendLoad(t *testing.T) {
 
 func TestMockBackendUnload(t *testing.T) {
 	b := NewMockBackend()
-	b.Load("/cognitiveos/models/test.gguf", nil)
+	_, _ = b.Load("/cognitiveos/models/test.gguf", nil)
 	if err := b.Unload(); err != nil {
 		t.Fatalf("unload: %v", err)
 	}
@@ -38,7 +38,7 @@ func TestMockBackendUnload(t *testing.T) {
 
 func TestMockBackendClose(t *testing.T) {
 	b := NewMockBackend()
-	b.Load("/cognitiveos/models/test.gguf", nil)
+	_, _ = b.Load("/cognitiveos/models/test.gguf", nil)
 	if err := b.Close(); err != nil {
 		t.Fatalf("close: %v", err)
 	}
@@ -57,7 +57,7 @@ func TestMockBackendGenerateWithoutLoad(t *testing.T) {
 
 func TestMockBackendGenerate(t *testing.T) {
 	b := NewMockBackend()
-	b.Load("/cognitiveos/models/test.gguf", nil)
+	_, _ = b.Load("/cognitiveos/models/test.gguf", nil)
 
 	resp, err := b.Generate(GenerateReq{Prompt: "What is 2+2?"}, nil)
 	if err != nil {
@@ -79,7 +79,7 @@ func TestMockBackendGenerate(t *testing.T) {
 
 func TestMockBackendGenerateStreaming(t *testing.T) {
 	b := NewMockBackend()
-	b.Load("/cognitiveos/models/test.gguf", nil)
+	_, _ = b.Load("/cognitiveos/models/test.gguf", nil)
 
 	var tokens []string
 	onToken := func(tok string) {

internal/model/model.go

@@ -97,7 +97,7 @@ func (m *Manager) Resolve(name string) (string, error) {
 
 	// Search recursively
 	var found string
-	filepath.Walk(m.ModelDir, func(path string, info os.FileInfo, err error) error {
+	_ = filepath.Walk(m.ModelDir, func(path string, info os.FileInfo, err error) error {
 		if err != nil || found != "" {
 			return nil
 		}

internal/model/model_test.go

@@ -9,7 +9,7 @@ import (
 func TestManagerResolveAbsPath(t *testing.T) {
 	dir := t.TempDir()
 	modelPath := filepath.Join(dir, "test.gguf")
-	os.WriteFile(modelPath, []byte("dummy"), 0644)
+	_ = os.WriteFile(modelPath, []byte("dummy"), 0644)
 
 	m := NewManager(dir)
 	got, err := m.Resolve(modelPath)
@@ -31,9 +31,9 @@ func TestManagerResolveNotFound(t *testing.T) {
 
 func TestManagerList(t *testing.T) {
 	dir := t.TempDir()
-	os.MkdirAll(filepath.Join(dir, "wide", "active"), 0755)
-	os.WriteFile(filepath.Join(dir, "raw-model.gguf"), []byte("raw"), 0644)
-	os.WriteFile(filepath.Join(dir, "wide", "active", "gemma.gguf"), []byte("gemma"), 0644)
+	_ = os.MkdirAll(filepath.Join(dir, "wide", "active"), 0755)
+	_ = os.WriteFile(filepath.Join(dir, "raw-model.gguf"), []byte("raw"), 0644)
+	_ = os.WriteFile(filepath.Join(dir, "wide", "active", "gemma.gguf"), []byte("gemma"), 0644)
 
 	m := NewManager(dir)
 	models, err := m.List()

internal/server/server.go

@@ -231,7 +231,7 @@ func (s *Server) handleGenerate(w http.ResponseWriter, r *http.Request) {
 
 		onToken := func(token string) {
 			line, _ := json.Marshal(llm.GenerateResp{Response: token, Done: false})
-			fmt.Fprintf(w, "%s\n", line)
+			_, _ = fmt.Fprintf(w, "%s\n", line)
 			flusher.Flush()
 		}
 
@@ -242,7 +242,7 @@ func (s *Server) handleGenerate(w http.ResponseWriter, r *http.Request) {
 		}
 		resp.Done = true
 		line, _ := json.Marshal(resp)
-		fmt.Fprintf(w, "%s\n", line)
+		_, _ = fmt.Fprintf(w, "%s\n", line)
 		flusher.Flush()
 		return
 	}
@@ -273,7 +273,7 @@ func (s *Server) handleChat(w http.ResponseWriter, r *http.Request) {
 
 	var promptBuilder strings.Builder
 	for _, msg := range req.Messages {
-		promptBuilder.WriteString(fmt.Sprintf("%s: %s\n", msg.Role, msg.Content))
+		fmt.Fprintf(&promptBuilder, "%s: %s\n", msg.Role, msg.Content)
 	}
 	prompt := promptBuilder.String()
 
@@ -312,7 +312,7 @@ func (s *Server) handleChat(w http.ResponseWriter, r *http.Request) {
 
 		onToken := func(token string) {
 			line, _ := json.Marshal(llm.GenerateResp{Response: token, Done: false})
-			fmt.Fprintf(w, "%s\n", line)
+			_, _ = fmt.Fprintf(w, "%s\n", line)
 			flusher.Flush()
 		}
 
@@ -323,7 +323,7 @@ func (s *Server) handleChat(w http.ResponseWriter, r *http.Request) {
 		}
 		resp.Done = true
 		line, _ := json.Marshal(resp)
-		fmt.Fprintf(w, "%s\n", line)
+		_, _ = fmt.Fprintf(w, "%s\n", line)
 		flusher.Flush()
 		return
 	}
@@ -444,7 +444,7 @@ func (s *Server) handleDelete(w http.ResponseWriter, r *http.Request) {
 
 	var req deleteRequest
 	if r.Method == http.MethodPost {
-		json.NewDecoder(r.Body).Decode(&req)
+		_ = json.NewDecoder(r.Body).Decode(&req)
 	}
 
 	s.mu.Lock()