diff --git a/client/main.go b/client/main.go
index 2a59a45..bf947f7 100644
--- a/client/main.go
+++ b/client/main.go
@@ -69,10 +69,15 @@ var rootCmd = &cobra.Command{
 		chunkCfg := cfg.ResolveChunking()
 
 		var embedder embed.Embedder
-		if embCfg.Model != "" {
-			embedder = embed.NewOpenAI(embCfg.BaseURL, embCfg.APIKey, embCfg.Model, embCfg.Dimension)
-		} else {
-			embedder = embed.NewBGE(embCfg.BaseURL, embCfg.APIKey, embCfg.Dimension)
+		switch embCfg.Provider {
+		case "allama":
+			embedder = embed.NewAllama(embCfg.BaseURL, embCfg.Model, embCfg.Dimension)
+		default:
+			if embCfg.Model != "" {
+				embedder = embed.NewOpenAI(embCfg.BaseURL, embCfg.APIKey, embCfg.Model, embCfg.Dimension)
+			} else {
+				embedder = embed.NewBGE(embCfg.BaseURL, embCfg.APIKey, embCfg.Dimension)
+			}
 		}
 
 		baseURL := os.Getenv("SERVER_URL")
diff --git a/internal/rag/config/runtime.go b/internal/rag/config/runtime.go
index 120fc21..2976a54 100644
--- a/internal/rag/config/runtime.go
+++ b/internal/rag/config/runtime.go
@@ -10,6 +10,7 @@ import (
 
 // RuntimeEmbedding is the resolved embedding configuration used at runtime.
 type RuntimeEmbedding struct {
+	Provider     string
 	BaseURL      string
 	APIKey       string
 	Model        string
@@ -23,6 +24,7 @@ type RuntimeEmbedding struct {
 func (c *Config) ResolveEmbedding() RuntimeEmbedding {
 	e := c.Embedding
 	var rt RuntimeEmbedding
+	rt.Provider = e.Provider
 	rt.Model = e.Model
 	rt.Dimension = e.Dimension
 	rt.RateLimitTPM = e.RateLimitTPM
@@ -83,8 +85,10 @@ type Runtime struct {
 	Datasources []DataSource `yaml:"datasources"`
 	Proxy       string       `yaml:"proxy"`
 	Embedding   struct {
+		Provider  string `yaml:"provider"`
 		BaseURL   string `yaml:"base_url"`
 		Token     string `yaml:"token"`
+		Model     string `yaml:"model"`
 		Dimension int    `yaml:"dimension"`
 	} `yaml:"embedding"`
 }
@@ -117,8 +121,10 @@ func (rt *Runtime) ToConfig() *Config {
 	c.Global.VectorDB = rt.VectorDB
 	c.Global.Datasources = rt.Datasources
 	c.Global.Proxy = rt.Proxy
+	c.Embedding.Provider = rt.Embedding.Provider
 	c.Embedding.BaseURL = rt.Embedding.BaseURL
 	c.Embedding.Token = rt.Embedding.Token
+	c.Embedding.Model = rt.Embedding.Model
 	c.Embedding.Dimension = rt.Embedding.Dimension
 	return &c
 }
diff --git a/internal/rag/embed/allama.go b/internal/rag/embed/allama.go
new file mode 100644
index 0000000..592e415
--- /dev/null
+++ b/internal/rag/embed/allama.go
@@ -0,0 +1,76 @@
+package embed
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+)
+
+// Allama implements the Embedder interface using the Allama/Ollama embeddings API.
+type Allama struct {
+	baseURL string
+	model   string
+	dim     int
+	client  *http.Client
+}
+
+// NewAllama creates a new Allama embedder.
+func NewAllama(baseURL, model string, dim int) *Allama {
+	return &Allama{
+		baseURL: strings.TrimRight(baseURL, "/"),
+		model:   model,
+		dim:     dim,
+		client:  &http.Client{Timeout: 30 * time.Second},
+	}
+}
+
+// Dimension returns the embedding dimension if known.
+func (a *Allama) Dimension() int { return a.dim }
+
+// Embed posts texts to the Allama embeddings endpoint.
+func (a *Allama) Embed(ctx context.Context, inputs []string) ([][]float32, int, error) {
+	vecs := make([][]float32, len(inputs))
+	url := a.baseURL + "/api/embeddings"
+	for i, text := range inputs {
+		payload := map[string]any{"model": a.model, "prompt": text}
+		body, _ := json.Marshal(payload)
+		req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+		if err != nil {
+			return nil, 0, err
+		}
+		req.Header.Set("Content-Type", "application/json")
+		resp, err := a.client.Do(req)
+		if err != nil {
+			return nil, 0, err
+		}
+		if resp.StatusCode >= 300 {
+			resp.Body.Close()
+			return nil, 0, fmt.Errorf("embed failed: %s", resp.Status)
+		}
+		data, err := io.ReadAll(resp.Body)
+		resp.Body.Close()
+		if err != nil {
+			return nil, 0, err
+		}
+		var out struct {
+			Embedding []float64 `json:"embedding"`
+		}
+		if err := json.Unmarshal(data, &out); err != nil {
+			return nil, 0, err
+		}
+		if a.dim == 0 {
+			a.dim = len(out.Embedding)
+		}
+		vec := make([]float32, len(out.Embedding))
+		for j, v := range out.Embedding {
+			vec[j] = float32(v)
+		}
+		vecs[i] = vec
+	}
+	return vecs, 0, nil
+}
diff --git a/internal/rag/ingest/ingest.go b/internal/rag/ingest/ingest.go
index 847f8ec..a015c58 100644
--- a/internal/rag/ingest/ingest.go
+++ b/internal/rag/ingest/ingest.go
@@ -64,10 +64,15 @@ func IngestRepo(ctx context.Context, cfg *cfgpkg.Config, ds cfgpkg.DataSource, o
 	defer conn.Close(ctx)
 
 	var embedder embed.Embedder
-	if embCfg.Model != "" {
-		embedder = embed.NewOpenAI(embCfg.BaseURL, embCfg.APIKey, embCfg.Model, embCfg.Dimension)
-	} else {
-		embedder = embed.NewBGE(embCfg.BaseURL, embCfg.APIKey, embCfg.Dimension)
+	switch embCfg.Provider {
+	case "allama":
+		embedder = embed.NewAllama(embCfg.BaseURL, embCfg.Model, embCfg.Dimension)
+	default:
+		if embCfg.Model != "" {
+			embedder = embed.NewOpenAI(embCfg.BaseURL, embCfg.APIKey, embCfg.Model, embCfg.Dimension)
+		} else {
+			embedder = embed.NewBGE(embCfg.BaseURL, embCfg.APIKey, embCfg.Dimension)
+		}
 	}
 	if err := store.EnsureSchema(ctx, conn, embedder.Dimension(), opt.MigrateDim); err != nil {
 		st.Errors = append(st.Errors, err)
diff --git a/internal/rag/service.go b/internal/rag/service.go
index e761766..2cce100 100644
--- a/internal/rag/service.go
+++ b/internal/rag/service.go
@@ -60,10 +60,15 @@ func (s *Service) Query(ctx context.Context, question string, limit int) ([]Docu
 		return nil, nil
 	}
 	var emb embed.Embedder
-	if embCfg.Model != "" {
-		emb = embed.NewOpenAI(embCfg.BaseURL, embCfg.APIKey, embCfg.Model, embCfg.Dimension)
-	} else {
-		emb = embed.NewBGE(embCfg.BaseURL, embCfg.APIKey, embCfg.Dimension)
+	switch embCfg.Provider {
+	case "allama":
+		emb = embed.NewAllama(embCfg.BaseURL, embCfg.Model, embCfg.Dimension)
+	default:
+		if embCfg.Model != "" {
+			emb = embed.NewOpenAI(embCfg.BaseURL, embCfg.APIKey, embCfg.Model, embCfg.Dimension)
+		} else {
+			emb = embed.NewBGE(embCfg.BaseURL, embCfg.APIKey, embCfg.Dimension)
+		}
 	}
 	vecs, _, err := emb.Embed(ctx, []string{question})
 	if err != nil {
diff --git a/server/api/askai.go b/server/api/askai.go
index 2f3db11..5b4fcc6 100644
--- a/server/api/askai.go
+++ b/server/api/askai.go
@@ -17,7 +17,7 @@ import (
 )
 
 // askFn performs the chat completion request. It is replaceable in tests.
-var askFn = callChutes
+var askFn = callLLM
 
 // registerAskAIRoutes wires the /api/askai endpoint.
 func registerAskAIRoutes(r *gin.RouterGroup) {
@@ -61,9 +61,10 @@ type serverConfig struct {
 	} `yaml:"api"`
 }
 
-// loadConfig reads model, URL, timeout and retries from ConfigPath and
-// environment variables. The Chutes API token is sourced from the config file.
-func loadConfig() (string, string, string, time.Duration, int) {
+// loadConfig reads provider, model, URL, timeout and retries from ConfigPath
+// and environment variables.
+func loadConfig() (string, string, string, string, time.Duration, int) {
+	provider := ""
 	model := os.Getenv("CHUTES_API_MODEL")
 	baseURL := os.Getenv("CHUTES_API_URL")
 	token := ""
@@ -74,19 +75,28 @@ func loadConfig() (string, string, string, time.Duration, int) {
 		var cfg serverConfig
 		if err := yaml.Unmarshal(data, &cfg); err == nil {
 			for _, p := range cfg.Provider {
-				if p.Name != "chutes" {
-					continue
+				if provider == "" {
+					provider = p.Name
 				}
-				if token == "" {
-					token = p.Token
+				switch p.Name {
+				case "allama":
+					if model == "" && len(p.Models) > 0 {
+						model = p.Models[0]
+					}
+					if baseURL == "" {
+						baseURL = p.BaseURL
+					}
+				case "chutes":
+					if token == "" {
+						token = p.Token
+					}
+					if model == "" && len(p.Models) > 0 {
+						model = p.Models[0]
+					}
+					if baseURL == "" {
+						baseURL = p.BaseURL
+					}
 				}
-				if model == "" && len(p.Models) > 0 {
-					model = p.Models[0]
-				}
-				if baseURL == "" {
-					baseURL = p.BaseURL
-				}
-				break
 			}
 			if cfg.API.AskAI.Timeout > 0 {
 				timeout = time.Duration(cfg.API.AskAI.Timeout) * time.Second
@@ -102,20 +112,28 @@ func loadConfig() (string, string, string, time.Duration, int) {
 	if retries > 3 {
 		retries = 3
 	}
-	if model == "" {
-		model = "deepseek-ai/DeepSeek-R1"
-	}
+	provider = strings.ToLower(provider)
 	baseURL = strings.TrimRight(baseURL, "/")
+	if provider == "allama" {
+		if baseURL == "" {
+			baseURL = "http://localhost:11434"
+		}
+		if model == "" {
+			model = "gpt-oss:20b"
+		}
+		return provider, token, model, baseURL + "/api/chat", timeout, retries
+	}
 	if baseURL == "" {
 		baseURL = "https://llm.chutes.ai"
 	}
-	url := baseURL + "/v1/chat/completions"
-	return token, model, url, timeout, retries
+	if model == "" {
+		model = "deepseek-ai/DeepSeek-R1"
+	}
+	return "chutes", token, model, baseURL + "/v1/chat/completions", timeout, retries
 }
 
 // callChutes sends the question to the hosted LLM service and returns the reply.
-func callChutes(question string) (string, error) {
-	token, model, url, timeout, retries := loadConfig()
+func callChutes(token, model, url string, timeout time.Duration, retries int, question string) (string, error) {
 	if token == "" || token == "cpk_xxxxxxx" {
 		return "", errors.New("chutes token not set")
 	}
@@ -181,3 +199,65 @@ func callChutes(question string) (string, error) {
 	}
 	return "", lastErr
 }
+
+// callAllama sends the question to a local Allama server.
+func callAllama(model, url string, timeout time.Duration, retries int, question string) (string, error) {
+	reqBody := map[string]any{
+		"model":    model,
+		"messages": []any{map[string]string{"role": "user", "content": question}},
+		"stream":   false,
+	}
+	data, err := json.Marshal(reqBody)
+	if err != nil {
+		return "", err
+	}
+	client := &http.Client{Timeout: timeout}
+	var lastErr error
+	for i := 0; i <= retries; i++ {
+		req, err := http.NewRequest("POST", url, bytes.NewReader(data))
+		if err != nil {
+			return "", err
+		}
+		req.Header.Set("Content-Type", "application/json")
+		resp, err := client.Do(req)
+		if err != nil {
+			lastErr = err
+			continue
+		}
+		b, err := io.ReadAll(resp.Body)
+		resp.Body.Close()
+		if err != nil {
+			lastErr = err
+			continue
+		}
+		if resp.StatusCode != http.StatusOK {
+			lastErr = fmt.Errorf("allama API error: %s", string(b))
+			continue
+		}
+		var res struct {
+			Message struct {
+				Content string `json:"content"`
+			} `json:"message"`
+		}
+		if err := json.Unmarshal(b, &res); err != nil {
+			lastErr = err
+			continue
+		}
+		return res.Message.Content, nil
+	}
+	if lastErr == nil {
+		lastErr = errors.New("request failed")
+	}
+	return "", lastErr
+}
+
+// callLLM dispatches the question to the configured provider.
+func callLLM(question string) (string, error) {
+	provider, token, model, url, timeout, retries := loadConfig()
+	switch provider {
+	case "allama":
+		return callAllama(model, url, timeout, retries, question)
+	default:
+		return callChutes(token, model, url, timeout, retries, question)
+	}
+}
diff --git a/server/api/rag_test.go b/server/api/rag_test.go
index 68bf6e9..438a531 100644
--- a/server/api/rag_test.go
+++ b/server/api/rag_test.go
@@ -120,7 +120,7 @@ func TestRAGUpsert_DimensionMismatch(t *testing.T) {
 	req.Header.Set("Content-Type", "application/json")
 	w := httptest.NewRecorder()
 	r.ServeHTTP(w, req)
-	if w.Code != http.StatusInternalServerError {
-		t.Fatalf("expected status 500, got %d", w.Code)
+	if w.Code != http.StatusServiceUnavailable {
+		t.Fatalf("expected status 503, got %d", w.Code)
 	}
 }
diff --git a/server/config/config.go b/server/config/config.go
index 8d99658..40afa47 100644
--- a/server/config/config.go
+++ b/server/config/config.go
@@ -59,6 +59,13 @@ type Global struct {
 	VectorDB    VectorDB     `yaml:"vectordb"`
 	Datasources []Datasource `yaml:"datasources"`
 	Proxy       string       `yaml:"proxy"`
+	Embedding   struct {
+		Provider  string `yaml:"provider"`
+		BaseURL   string `yaml:"base_url"`
+		Token     string `yaml:"token"`
+		Model     string `yaml:"model"`
+		Dimension int    `yaml:"dimension"`
+	} `yaml:"embedding"`
 }
 
 type Provider struct {
diff --git a/server/config/server.yaml b/server/config/server.yaml
index be23c9d..f5c61c8 100644
--- a/server/config/server.yaml
+++ b/server/config/server.yaml
@@ -1,15 +1,25 @@
 log:
   level: info
 global:
+  #proxy: socks5://127.0.0.1:1080 # optional
   redis:
     addr: 127.0.0.1:6379
+    password: ""
   vectordb:
     pgurl: postgres://user:password@127.0.0.1:5432/postgres
   datasources: []
   embedding:
-    base_url: http://127.0.0.1:11434
+    provider: allama
+    base_url: http://localhost:11434
     token: ""
+    model: bge-m3
     dimension: 1536
+provider:
+  - name: allama
+    base_url: http://localhost:11434
+    token: ""
+    models:
+      - gpt-oss:20b
 api:
   askai:
     timeout: 100