Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions changelog.d/fixed-15032-empty-inference-model.md
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
- Inference requests inherit a configured gateway default when the agent model is blank; if both are blank, Hive rejects the request locally and shows an actionable model configuration error instead of repeatedly sending an empty model to the provider ([#15032](https://github.com/hivecommons/hive/issues/15032)).
5 changes: 5 additions & 0 deletions src/crux/proxy/anthropic_translate.go
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,11 @@ import (
// open-source models toward agentic tool-calling behaviour. Pass "" to
// leave the system prompt unchanged (used by tests and non-inference paths).
func translateAnthropicToOpenAI(body []byte, targetModel string, maxContextLen int, preamble string) ([]byte, error) {
// The CLI's model is a Claude alias, not necessarily a gateway model ID.
// Never use it as a fallback or send an empty model to the provider.
if strings.TrimSpace(targetModel) == "" {
return nil, fmt.Errorf("no inference model configured: select an agent model or set the gateway default_model in Model Gateways")
}
var req anthropicRequest
if err := json.Unmarshal(body, &req); err != nil {
return nil, fmt.Errorf("unmarshal anthropic request: %w", err)
Expand Down
12 changes: 11 additions & 1 deletion src/crux/proxy/github_proxy_inference.go
Original file line number Diff line number Diff line change
Expand Up @@ -243,6 +243,14 @@ func (p *GitHubProxy) recordInferenceError(route *InferenceRoute, agentName stri
// and concurrent requests race on nothing. Unknown entitlements, verbatim
// matches, and ambiguous candidates all pass the route through unchanged.
func (p *GitHubProxy) routeWithEntitledModel(route *InferenceRoute, agentName string) *InferenceRoute {
// A blank per-agent selection inherits the configured backend default,
// even before an entitlement probe has completed. Keep the stored snapshot
// untouched, as with entitlement rewrites below.
if route != nil && strings.TrimSpace(route.Model) == "" && strings.TrimSpace(route.DefaultModel) != "" {
rewritten := *route
rewritten.Model = route.DefaultModel
route = &rewritten
}
if route == nil || route.Backend != "litellm" || p.entitlements == nil {
return route
}
Expand Down Expand Up @@ -429,7 +437,8 @@ func (p *GitHubProxy) inferenceTranslatorHandler() http.Handler {
openaiBody, err := translateAnthropicToOpenAI(body, route.Model, route.MaxContextLen, resolveInferencePreamble(route, agentName))
if err != nil {
p.logger.Error("inference translate request failed", "agent", agentName, "error", err)
http.Error(w, fmt.Sprintf(`{"type":"error","error":{"type":"api_error","message":"translation error: %s"}}`, err.Error()), http.StatusBadGateway)
p.recordAgentInferenceError(agentName, "translation error: "+err.Error(), 0)
http.Error(w, fmt.Sprintf(`{"type":"error","error":{"type":"api_error","message":"translation error: %s"}}`, jsonEscape(err.Error())), http.StatusBadGateway)
return
}

Expand Down Expand Up @@ -645,6 +654,7 @@ func (p *GitHubProxy) handleInferenceRequest(conn net.Conn, req *http.Request, a
openaiBody, err := translateAnthropicToOpenAI(body, route.Model, route.MaxContextLen, resolveInferencePreamble(route, agentName))
if err != nil {
p.logger.Error("inference translate request failed", "agent", agentName, "error", err)
p.recordAgentInferenceError(agentName, "translation error: "+err.Error(), 0)
p.writeHTTPError(conn, http.StatusBadGateway, "translation error: "+err.Error())
return
}
Expand Down
103 changes: 103 additions & 0 deletions src/crux/proxy/inference_missing_model_test.go
Original file line number Diff line number Diff line change
@@ -0,0 +1,103 @@
package proxy

import (
"bufio"
"encoding/json"
"io"
"net"
"net/http"
"net/http/httptest"
"strings"
"sync/atomic"
"testing"
)

// Both entry points must keep blank models away from the provider, expose the
// configuration remedy, and recover when a gateway default is supplied.
func TestInferenceMissingModel(t *testing.T) {
for _, entry := range []string{"translator", "mitm"} {
for _, tc := range []struct {
name string
model string
defaultModel string
wantModel string
}{
{name: "empty"},
{name: "whitespace", model: " \t", defaultModel: " \n"},
{name: "gateway default", defaultModel: "Azure/gpt-4o", wantModel: "Azure/gpt-4o"},
{name: "whitespace inherits default", model: " \t", defaultModel: "Azure/gpt-4o", wantModel: "Azure/gpt-4o"},
{name: "explicit selection", model: "aws/claude-sonnet-4-6", defaultModel: "Azure/gpt-4o", wantModel: "aws/claude-sonnet-4-6"},
} {
t.Run(entry+"/"+tc.name, func(t *testing.T) {
var hits atomic.Int32
upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
hits.Add(1)
var req openaiRequest
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
t.Errorf("decode upstream request: %v", err)
}
if req.Model == "" || req.Model != tc.wantModel {
t.Errorf("upstream model = %q, want %q", req.Model, tc.wantModel)
}
w.Header().Set("Content-Type", "application/json")
io.WriteString(w, `{"id":"test","object":"chat.completion","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}]}`)
}))
defer upstream.Close()
p := newInferenceTestProxy()
p.inferenceLastErrors = newInferenceLastErrorState()
route := &InferenceRoute{Backend: "litellm", Endpoint: upstream.URL, Model: tc.model, DefaultModel: tc.defaultModel}
const body = `{"model":"claude-cli-alias","max_tokens":8,"messages":[{"role":"user","content":"hi"}]}`
var resp *http.Response
if entry == "translator" {
p.inference.Set("scanner", route)
req := httptest.NewRequest(http.MethodPost, "/v1/messages", strings.NewReader(body))
req.Header.Set("x-api-key", "sk-hive-scanner")
w := httptest.NewRecorder()
p.inferenceTranslatorHandler().ServeHTTP(w, req)
resp = w.Result()
} else {
client, server := net.Pipe()
defer client.Close()
done := make(chan struct{})
go func() {
defer close(done)
defer server.Close()
req, _ := http.NewRequest(http.MethodPost, "https://api.anthropic.com/v1/messages", strings.NewReader(body))
p.handleInferenceRequest(server, req, "scanner", route)
}()
var err error
resp, err = http.ReadResponse(bufio.NewReader(client), nil)
if err != nil {
t.Fatal(err)
}
defer func() { client.Close(); <-done }()
}
defer resp.Body.Close()
responseBody, err := io.ReadAll(resp.Body)
if err != nil {
t.Fatal(err)
}
if !json.Valid(responseBody) {
t.Fatalf("invalid response JSON: %s", responseBody)
}
if tc.wantModel == "" {
if resp.StatusCode != http.StatusBadGateway || !strings.Contains(string(responseBody), "no inference model configured") || !strings.Contains(string(responseBody), "default_model") {
t.Fatalf("missing configuration response: %d %s", resp.StatusCode, responseBody)
}
if hits.Load() != 0 {
t.Fatal("missing model was forwarded upstream")
}
msg, status, _, ok := p.LastInferenceError("scanner")
if !ok || status != 0 || !strings.Contains(msg, "default_model") {
t.Fatalf("missing dashboard error: %q, %d, %v", msg, status, ok)
}
} else if resp.StatusCode != http.StatusOK || hits.Load() != 1 {
t.Fatalf("configured request: status=%d hits=%d body=%s", resp.StatusCode, hits.Load(), responseBody)
}
if route.Model != tc.model {
t.Fatalf("stored model mutated: %q, want %q", route.Model, tc.model)
}
})
}
}
}
Loading