completion-kit 0.5.35 → 0.5.36

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 7ec9f216056f47d007b8a512a009dea02aeaad328a9c4832ac0c1e122b816b15
-  data.tar.gz: 413f1b8e8ca28ed2c14e55210299a28bef42acb1a437a3ee9c3bd88b62d03bf9
+  metadata.gz: f3189fdba715e4750befcadf517462d0523ae7366611964388309031ad6a4d4f
+  data.tar.gz: 322e2dd4847c5e8bd3b8af83b4b09873efcd4238c945d928fdac7c7ffbbaf7de
 SHA512:
-  metadata.gz: c837fd6ddc33c5799145bac2a0b1dba4ca1df807365c621330b619c47607e666a30325b47ce6d34e2ecb93e13d4e14a9538d597e0d56e47b9f7a1a836432be0d
-  data.tar.gz: 06b9ac7200883cd19d11ff127898d52d940b3f7a110cd5e7744bfcb2abab797fa05e923ba76121f6c00c6aeb8ac5d7b5d5e8b79b07d773ad0c0d5650ec41b90d
+  metadata.gz: f611cfbc07196fd75eb16962bb1acf4a271d759f42473af9b94d049860839a12657223c72bb12aaeaa80a14adfbb29a20c4cb3c4c3176193ce44966bb876b011
+  data.tar.gz: 2806d017ce92c625e6c7f83e789d13afe53065107a5a803f0d309cb1eaee65d752a8b1e18734c42e9106fc6c38c2cbbfcc803f76477571d7e2330130ebe8eee1

data/app/services/completion_kit/model_discovery_service.rb

@@ -106,14 +106,18 @@ module CompletionKit
 
     def fetch_ollama_models
       raise DiscoveryError, "Ollama endpoint URL is required" if @api_endpoint.blank?
-      base_url = @api_endpoint.to_s.delete_suffix("/")
-      response = fetch_connection(base_url).get("/models") do |req|
+      base_url = ollama_root_url
+      response = fetch_connection(base_url).get("/v1/models") do |req|
         req.headers["Authorization"] = "Bearer #{@api_key}" if @api_key.present?
       end
       raise_fetch_error!(response) unless response.success?
       JSON.parse(response.body).fetch("data", []).map { |e| { id: e["id"], display_name: e["id"] } }
     end
 
+    def ollama_root_url
+      @api_endpoint.to_s.strip.delete_suffix("/").delete_suffix("/v1")
+    end
+
     def reconcile(discovered)
       api_model_ids = discovered.map { |m| m[:id] }
       meta_by_id = discovered.index_by { |m| m[:id] }
@@ -336,15 +340,14 @@ module CompletionKit
     end
 
     def ollama_probe(model_id, input, max_tokens)
-      base_url = @api_endpoint.to_s.delete_suffix("/")
-      conn = Faraday.new(url: base_url) do |f|
+      conn = Faraday.new(url: ollama_root_url) do |f|
         f.options.timeout = 60
         f.options.open_timeout = 5
         f.request :retry, max: 1, interval: 0.5
         f.adapter Faraday.default_adapter
       end
       conn.post do |req|
-        req.url "/chat/completions"
+        req.url "/v1/chat/completions"
         req.headers["Content-Type"] = "application/json"
         req.headers["Authorization"] = "Bearer #{@api_key}" if @api_key.present?
         req.body = { model: model_id, messages: [{ role: "user", content: input }], max_tokens: max_tokens }.to_json

data/lib/completion_kit/version.rb

@@ -1,3 +1,3 @@
 module CompletionKit
-  VERSION = "0.5.35"
+  VERSION = "0.5.36"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: completion-kit
 version: !ruby/object:Gem::Version
-  version: 0.5.35
+  version: 0.5.36
 platform: ruby
 authors:
 - Damien Bastin