LCORE-4692: add TieredCompaction builder and pydantic-ai-harness dependency

.konflux/requirements-build.txt

@@ -42,6 +42,7 @@ hatchling==1.33.0
     #   logfire
     #   logfire-sdk
     #   pydantic-ai
+    #   pydantic-ai-harness
     #   pydantic-ai-skills
     #   pydantic-ai-slim
     #   pydantic-evals
@@ -95,6 +96,7 @@ uv-dynamic-versioning==0.14.1
     # via
     #   genai-prices
     #   pydantic-ai
+    #   pydantic-ai-harness
     #   pydantic-ai-slim
     #   pydantic-evals
     #   pydantic-graph

.konflux/requirements.hashes.source.txt

@@ -41,6 +41,9 @@ presidio-anonymizer==2.2.366 \
 pydantic-ai==2.38.0 \
     --hash=sha256:2dc423fefdaf04998889440404ec84ea9060f7bc9dfadbc94be1f5c8cbcc0324 \
     --hash=sha256:90fff4dd3b1db9994861bd102c786d471b81e07a3165c751f403b414bd02e3f9
+pydantic-ai-harness==0.30.0 \
+    --hash=sha256:226d35fc3602852e92adf26b5a29717721d66da0f99d5cbff6043bf49ecb0fee \
+    --hash=sha256:f9eaa6c19abf47e7af9a84c86124e96c59b432afe8b9f7deae3d9e98fa4677ef
 pydantic-ai-skills==1.4.0 \
     --hash=sha256:7303e0738a837218415f8b2e7bd5b5cbc0ceeb74850b1902c2d079550e0b9b83 \
     --hash=sha256:bce8731a042f50c45965acc520dc883163e511c34b208f720eb62067ef5be686

docs/devel_doc/openapi.json

@@ -13417,12 +13417,24 @@
                         "title": "Buffer max ratio",
                         "description": "Maximum fraction of context window the buffer zone can occupy, regardless of buffer_turns.",
                         "default": 0.3
+                    },
+                    "summary_prompt": {
+                        "anyOf": [
+                            {
+                                "type": "string"
+                            },
+                            {
+                                "type": "null"
+                            }
+                        ],
+                        "title": "Summary prompt",
+                        "description": "Prompt template for summarizing older turns in the Pydantic AI compaction. The conversation transcript replaces the {messages} placeholder, which is required; no other replacement field is allowed, so double any literal braces. When unset, the default prompt is used."
                     }
                 },
                 "additionalProperties": false,
                 "type": "object",
                 "title": "CompactionConfiguration",
-                "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard \u2014 if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns."
+                "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard \u2014 if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns.\n    summary_prompt: Optional prompt template for summarizing older\n        turns in the Pydantic AI compaction. It must contain the\n        ``{messages}`` placeholder and no other replacement field;\n        when unset, the default prompt is used."
             },
             "Configuration": {
                 "properties": {

docs/devel_doc/openapi.md

@@ -6078,6 +6078,10 @@ Attributes:
     buffer_max_ratio: Hard cap on the fraction of the context
         window the buffer zone may occupy, regardless of
         buffer_turns.
+    summary_prompt: Optional prompt template for summarizing older
+        turns in the Pydantic AI compaction. It must contain the
+        ``{messages}`` placeholder and no other replacement field;
+        when unset, the default prompt is used.
 
 
 | Field | Type | Description |
@@ -6087,6 +6091,7 @@ Attributes:
 | token_floor | integer | Minimum token count before compaction can trigger. Prevents triggering on very small context windows. |
 | buffer_turns | integer | Number of recent turns to keep verbatim. |
 | buffer_max_ratio | number | Maximum fraction of context window the buffer zone can occupy, regardless of buffer_turns. |
+| summary_prompt |  | Prompt template for summarizing older turns in the Pydantic AI compaction. The conversation transcript replaces the {messages} placeholder, which is required; no other replacement field is allowed, so double any literal braces. When unset, the default prompt is used. |
 
 
 ## Configuration

docs/models/successful_responses.json

@@ -868,7 +868,7 @@
             },
             "CompactionConfiguration": {
                 "additionalProperties": false,
-                "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard \u2014 if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns.",
+                "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard \u2014 if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns.\n    summary_prompt: Optional prompt template for summarizing older\n        turns in the Pydantic AI compaction. It must contain the\n        ``{messages}`` placeholder and no other replacement field;\n        when unset, the default prompt is used.",
                 "properties": {
                     "enabled": {
                         "default": false,
@@ -901,6 +901,13 @@
                         "description": "Maximum fraction of context window the buffer zone can occupy, regardless of buffer_turns.",
                         "title": "Buffer max ratio",
                         "type": "number"
+                    },
+                    "summary_prompt": {
+                        "type": "string",
+                        "nullable": true,
+                        "default": null,
+                        "description": "Prompt template for summarizing older turns in the Pydantic AI compaction. The conversation transcript replaces the {messages} placeholder, which is required; no other replacement field is allowed, so double any literal braces. When unset, the default prompt is used.",
+                        "title": "Summary prompt"
                     }
                 },
                 "title": "CompactionConfiguration",

docs/models/successful_responses.md

@@ -418,6 +418,10 @@ Attributes:
     buffer_max_ratio: Hard cap on the fraction of the context
         window the buffer zone may occupy, regardless of
         buffer_turns.
+    summary_prompt: Optional prompt template for summarizing older
+        turns in the Pydantic AI compaction. It must contain the
+        ``{messages}`` placeholder and no other replacement field;
+        when unset, the default prompt is used.
 
 
 | Field | Type | Description |
@@ -427,6 +431,7 @@ Attributes:
 | token_floor | integer | Minimum token count before compaction can trigger. Prevents triggering on very small context windows. |
 | buffer_turns | integer | Number of recent turns to keep verbatim. |
 | buffer_max_ratio | number | Maximum fraction of context window the buffer zone can occupy, regardless of buffer_turns. |
+| summary_prompt | string | Prompt template for summarizing older turns in the Pydantic AI compaction. The conversation transcript replaces the {messages} placeholder, which is required; no other replacement field is allowed, so double any literal braces. When unset, the default prompt is used. |
 
 
 ## Configuration

docs/user_doc/config.html

@@ -555,7 +555,10 @@ <h2 id="compactionconfiguration">CompactionConfiguration</h2>
 &#x2014; if these turns exceed the available budget, it reduces buffer_turns by
 one repeatedly until the budget fits, down to zero. buffer_max_ratio:
 Hard cap on the fraction of the context window the buffer zone may
-occupy, regardless of buffer_turns.</p>
+occupy, regardless of buffer_turns. summary_prompt: Optional prompt
+template for summarizing older turns in the Pydantic AI compaction. It
+must contain the <code>{messages}</code> placeholder and no other
+replacement field; when unset, the default prompt is used.</p>
     <table>
       <colgroup>
         <col style="width: 26%"/>
@@ -599,6 +602,15 @@ <h2 id="compactionconfiguration">CompactionConfiguration</h2>
           <td>Maximum fraction of context window the buffer zone can occupy,
 regardless of buffer_turns.</td>
         </tr>
+        <tr>
+          <td>summary_prompt</td>
+          <td>string</td>
+          <td>Prompt template for summarizing older turns in the Pydantic AI
+compaction. The conversation transcript replaces the {messages}
+placeholder, which is required; no other replacement field is allowed,
+so double any literal braces. When unset, the default prompt is
+used.</td>
+        </tr>
       </tbody>
     </table>
     <h2 id="configuration">Configuration</h2>

docs/user_doc/config.json

@@ -389,7 +389,7 @@
       },
       "CompactionConfiguration": {
         "additionalProperties": false,
-        "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard — if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns.",
+        "description": "Configuration for conversation history compaction.\n\nCompaction summarizes older conversation turns when their estimated\ntoken count approaches the context window limit, keeping the\nconversation usable instead of failing with HTTP 413. The\nconfiguration here controls when compaction triggers and how much\nrecent context is preserved verbatim.\n\nAttributes:\n    enabled: Master switch. When False, compaction never triggers\n        and other fields are inert.\n    threshold_ratio: Trigger compaction when estimated input tokens\n        exceed this fraction of the model's context window\n        (clamped to 0.0..1.0).\n    token_floor: Minimum estimated token count before compaction\n        can trigger, regardless of threshold_ratio. Prevents\n        triggering on very small context windows.\n    buffer_turns: Initial number of recent turns to keep verbatim.\n        The runtime applies a degrading guard — if these turns\n        exceed the available budget, it reduces buffer_turns by\n        one repeatedly until the budget fits, down to zero.\n    buffer_max_ratio: Hard cap on the fraction of the context\n        window the buffer zone may occupy, regardless of\n        buffer_turns.\n    summary_prompt: Optional prompt template for summarizing older\n        turns in the Pydantic AI compaction. It must contain the\n        ``{messages}`` placeholder and no other replacement field;\n        when unset, the default prompt is used.",
         "properties": {
           "enabled": {
             "default": false,
@@ -422,6 +422,13 @@
             "description": "Maximum fraction of context window the buffer zone can occupy, regardless of buffer_turns.",
             "title": "Buffer max ratio",
             "type": "number"
+          },
+          "summary_prompt": {
+            "type": "string",
+            "nullable": true,
+            "default": null,
+            "description": "Prompt template for summarizing older turns in the Pydantic AI compaction. The conversation transcript replaces the {messages} placeholder, which is required; no other replacement field is allowed, so double any literal braces. When unset, the default prompt is used.",
+            "title": "Summary prompt"
           }
         },
         "title": "CompactionConfiguration",

docs/user_doc/config.md

@@ -209,15 +209,20 @@ Attributes:
     buffer_max_ratio: Hard cap on the fraction of the context
         window the buffer zone may occupy, regardless of
         buffer_turns.
-
-
-| Field            | Type    | Description                                                                                                 |
-|------------------|---------|--------------                                                                                               |
-| enabled          | boolean | When true, older conversation turns are summarized when estimated tokens approach the context window limit. |
-| threshold_ratio  | number  | Trigger compaction when estimated tokens exceed this fraction of the model's context window (0.0-1.0).      |
-| token_floor      | integer | Minimum token count before compaction can trigger. Prevents triggering on very small context windows.       |
-| buffer_turns     | integer | Number of recent turns to keep verbatim.                                                                    |
-| buffer_max_ratio | number  | Maximum fraction of context window the buffer zone can occupy, regardless of buffer_turns.                  |
+    summary_prompt: Optional prompt template for summarizing older
+        turns in the Pydantic AI compaction. It must contain the
+        ``{messages}`` placeholder and no other replacement field;
+        when unset, the default prompt is used.
+
+
+| Field            | Type    | Description                                                                                                                                                                                                                                                                 |
+|------------------|---------|--------------                                                                                                                                                                                                                                                               |
+| enabled          | boolean | When true, older conversation turns are summarized when estimated tokens approach the context window limit.                                                                                                                                                                 |
+| threshold_ratio  | number  | Trigger compaction when estimated tokens exceed this fraction of the model's context window (0.0-1.0).                                                                                                                                                                      |
+| token_floor      | integer | Minimum token count before compaction can trigger. Prevents triggering on very small context windows.                                                                                                                                                                       |
+| buffer_turns     | integer | Number of recent turns to keep verbatim.                                                                                                                                                                                                                                    |
+| buffer_max_ratio | number  | Maximum fraction of context window the buffer zone can occupy, regardless of buffer_turns.                                                                                                                                                                                  |
+| summary_prompt   | string  | Prompt template for summarizing older turns in the Pydantic AI compaction. The conversation transcript replaces the {messages} placeholder, which is required; no other replacement field is allowed, so double any literal braces. When unset, the default prompt is used. |
 
 
 ## Configuration