Include the served manifest digest in local /api/chat responses (tested v0.34.0 patch)
Author: Belphareon-bakCreated Sep 11, 2026Updated Sep 17, 2026
Local evaluation clients need to identify the exact artifact that served a chat response. /api/tags before and after inference cannot distinguish an A -> B -> A tag replacement during the request. A model name alone is therefore insufficient to bind a score to an artifact.
Proposed additive response fields for local /api/chat:
digest: full local manifest digest from the same*Modelselected for this request, in streaming and non-streaming responses, covering native and template paths.provider_version: serving Ollama version, useful for keeping measurements from different inference-engine versions separate. This field could be considered separately if preferred.
The patch below is based on v0.34.0 (d8ab4b4f0ca24b51d3a46b3bf4f462e58ce66b1f). It adds no inventory polling or client lease requirement. Remote/cloud proxy responses are not relabeled as a locally served artifact.
Validation on Linux x86_64, RTX 3090, Go 1.26.7, with the matching v0.34.0 native payload:
go test ./server -run '^(TestGenerateChat|TestChatHandlerChatTemplateRoute)$' -count=1passes with an isolated model directory.- Phi-4 14B and Qwen3.5 27B return the expected full manifest digest and serving version in non-streaming responses and every streaming frame.
- Both models completed actual GPU inference; the local evaluation client rejects a missing/mismatched digest or serving version.
This is a tested implementation proposal, not a claim that a response field protects against an untrusted server. The intended boundary is accidental/concurrent artifact replacement on a trusted local provider.
Applicable git-am patch against v0.34.0From e8c2d86c3a056b58031c77ec5141a8ecd7cf923a Mon Sep 17 00:00:00 2001
From: Belphareon <[email protected]>
Date: Fri, 11 Sep 2026 21:00:00 +0000
Subject: [PATCH] server: attest chat manifest digest and serving provider
version
---
api/types.go | 6 ++++
server/routes.go | 60 ++++++++++++++++++++--------------
server/routes_generate_test.go | 45 +++++++++++++++++++++++--
3 files changed, 85 insertions(+), 26 deletions(-)
diff --git a/api/types.go b/api/types.go
index 7ea9022..beeeb85 100644
--- a/api/types.go
+++ b/api/types.go
@@ -521,6 +521,12 @@ type ChatResponse struct {
// Model is the model name that generated the response.
Model string `json:"model"`
+ // Digest is the exact local manifest digest selected for this response.
+ Digest string `json:"digest,omitempty"`
+
+ // ProviderVersion identifies the Ollama build serving this response.
+ ProviderVersion string `json:"provider_version,omitempty"`
+
// RemoteModel is the name of the upstream model that generated the response.
RemoteModel string `json:"remote_model,omitempty"`
diff --git a/server/routes.go b/server/routes.go
index 9b73139..eed328d 100644
--- a/server/routes.go
+++ b/server/routes.go
@@ -2520,11 +2520,13 @@ func (s *Server) ChatHandler(c *gin.Context) {
s.sched.expireRunner(m)
c.JSON(http.StatusOK, api.ChatResponse{
- Model: req.Model,
- CreatedAt: time.Now().UTC(),
- Message: api.Message{Role: "assistant"},
- Done: true,
- DoneReason: "unload",
+ Digest: m.Digest,
+ ProviderVersion: version.Version,
+ Model: req.Model,
+ CreatedAt: time.Now().UTC(),
+ Message: api.Message{Role: "assistant"},
+ Done: true,
+ DoneReason: "unload",
})
return
}
@@ -2658,11 +2660,13 @@ func (s *Server) ChatHandler(c *gin.Context) {
if len(req.Messages) == 0 {
c.JSON(http.StatusOK, api.ChatResponse{
- Model: req.Model,
- CreatedAt: time.Now().UTC(),
- Message: api.Message{Role: "assistant"},
- Done: true,
- DoneReason: "load",
+ Digest: m.Digest,
+ ProviderVersion: version.Version,
+ Model: req.Model,
+ CreatedAt: time.Now().UTC(),
+ Message: api.Message{Role: "assistant"},
+ Done: true,
+ DoneReason: "load",
})
return
}
@@ -2721,8 +2725,10 @@ func (s *Server) ChatHandler(c *gin.Context) {
// If debug mode is enabled, return the rendered template instead of calling the model
if req.DebugRenderOnly {
c.JSON(http.StatusOK, api.ChatResponse{
- Model: req.Model,
- CreatedAt: time.Now().UTC(),
+ Digest: m.Digest,
+ ProviderVersion: version.Version,
+ Model: req.Model,
+ CreatedAt: time.Now().UTC(),
DebugInfo: &api.DebugInfo{
RenderedTemplate: prompt,
ImageCount: len(media),
@@ -2823,12 +2829,14 @@ func (s *Server) ChatHandler(c *gin.Context) {
}
res := api.ChatResponse{
- Model: req.Model,
- CreatedAt: time.Now().UTC(),
- Message: api.Message{Role: "assistant", Content: r.Content},
- Done: r.Done,
- Metrics: metrics,
- Logprobs: toAPILogprobs(r.Logprobs),
+ Digest: m.Digest,
+ ProviderVersion: version.Version,
+ Model: req.Model,
+ CreatedAt: time.Now().UTC(),
+ Message: api.Message{Role: "assistant", Content: r.Content},
+ Done: r.Done,
+ Metrics: metrics,
+ Logprobs: toAPILogprobs(r.Logprobs),
}
if r.Done {
@@ -3015,8 +3023,10 @@ func (s *Server) handleNativeChat(c *gin.Context, req api.ChatRequest, m *Model,
}
c.JSON(http.StatusOK, api.ChatResponse{
- Model: req.Model,
- CreatedAt: time.Now().UTC(),
+ Digest: m.Digest,
+ ProviderVersion: version.Version,
+ Model: req.Model,
+ CreatedAt: time.Now().UTC(),
DebugInfo: &api.DebugInfo{
RenderedTemplate: prompt,
ImageCount: countChatImages(msgs),
@@ -3031,10 +3041,12 @@ func (s *Server) handleNativeChat(c *gin.Context, req api.ChatRequest, m *Model,
err := r.Chat(c.Request.Context(), nativeReq, func(r llm.ChatResponse) {
res := api.ChatResponse{
- Model: req.Model,
- CreatedAt: time.Now().UTC(),
- Message: r.Message,
- Done: r.Done,
+ Digest: m.Digest,
+ ProviderVersion: version.Version,
+ Model: req.Model,
+ CreatedAt: time.Now().UTC(),
+ Message: r.Message,
+ Done: r.Done,
Metrics: api.Metrics{
PromptEvalCount: r.PromptEvalCount,
PromptEvalCachedCount: r.PromptEvalCachedCount,
diff --git a/server/routes_generate_test.go b/server/routes_generate_test.go
index f3decb8..3f81d9a 100644
--- a/server/routes_generate_test.go
+++ b/server/routes_generate_test.go
@@ -4,6 +4,7 @@ import (
"bytes"
"context"
"encoding/json"
+ "github.com/ollama/ollama/version"
"io"
"net/http"
"net/http/httptest"
@@ -312,6 +313,20 @@ func TestChatHandlerChatTemplateRoute(t *testing.T) {
if err := json.Unmarshal(w.Body.Bytes(), &actual); err != nil {
t.Fatal(err)
}
+ existingName, err := getExistingName(model.ParseName("chat-template"))
+ if err != nil {
+ t.Fatal(err)
+ }
+ servedModel, err := GetModel(existingName.String())
+ if err != nil {
+ t.Fatal(err)
+ }
+ if actual.ProviderVersion != version.Version {
+ t.Fatalf("provider version = %q, want %q", actual.ProviderVersion, version.Version)
+ }
+ if actual.Digest != servedModel.Digest {
+ t.Fatalf("response digest = %q, want exact manifest digest %q", actual.Digest, servedModel.Digest)
+ }
if actual.Message.Content != "chat template response" {
t.Fatalf("expected chat template response, got %q", actual.Message.Content)
}
@@ -833,6 +848,18 @@ func TestGenerateChat(t *testing.T) {
if w.Code != http.StatusOK {
t.Fatalf("expected status 200, got %d", w.Code)
}
+ expectedDigest := func(t *testing.T, modelName string) string {
+ t.Helper()
+ existingName, err := getExistingName(model.ParseName(modelName))
+ if err != nil {
+ t.Fatal(err)
+ }
+ servedModel, err := GetModel(existingName.String())
+ if err != nil {
+ t.Fatal(err)
+ }
+ return servedModel.Digest
+ }
t.Run("missing body", func(t *testing.T) {
w := createRequest(t, s.ChatHandler, nil)
@@ -935,6 +962,13 @@ func TestGenerateChat(t *testing.T) {
if actual.Model != "test" {
t.Errorf("expected model test, got %s", actual.Model)
}
+ wantDigest := expectedDigest(t, "test")
+ if actual.ProviderVersion != version.Version {
+ t.Fatalf("provider version = %q, want %q", actual.ProviderVersion, version.Version)
+ }
+ if actual.Digest != wantDigest {
+ t.Errorf("response digest = %q, want exact manifest digest %q", actual.Digest, wantDigest)
+ }
if !actual.Done {
t.Errorf("expected done true, got false")
@@ -945,7 +979,7 @@ func TestGenerateChat(t *testing.T) {
}
})
- checkChatResponse := func(t *testing.T, body io.Reader, model, content string) {
+ checkChatResponse := func(t *testing.T, body io.Reader, modelName, content string) {
t.Helper()
var actual api.ChatResponse
@@ -953,9 +987,16 @@ func TestGenerateChat(t *testing.T) {
t.Fatal(err)
}
- if actual.Model != model {
+ if actual.Model != modelName {
t.Errorf("expected model test, got %s", actual.Model)
}
+ wantDigest := expectedDigest(t, modelName)
+ if actual.ProviderVersion != version.Version {
+ t.Fatalf("provider version = %q, want %q", actual.ProviderVersion, version.Version)
+ }
+ if actual.Digest != wantDigest {
+ t.Errorf("response digest = %q, want exact manifest digest %q", actual.Digest, wantDigest)
+ }
if !actual.Done {
t.Errorf("expected done false, got true")
--
2.43.0
Source: ollama/ollama