Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

add bot temperature (#2808)

DuTaoca29e5e4b7fc
diff --git a/bot/README.md b/bot/README.md
index 3676de79b..de27fd549 100644
--- a/bot/README.md
+++ b/bot/README.md
@@ -167,6 +167,7 @@ The bot will connect to the remote OpenViking server. Please start the OpenVikin
All configurations are under the `bot` field in `ov.conf`, with default values for configuration items. The optional manual configuration items are described as follows:
- `agents`: Agent configuration
- `model`: LLM model name used by the bot. When `provider` is set, use the provider-native model name (for example `doubao-seed-2-0-pro-260215`).
+ - `temperature`: Sampling temperature for LLM requests. Defaults to `0.7`.
- `provider`: Optional model provider name. When set, vikingbot uses OpenViking's `VLMFactory` + adapter path to create the backend directly (for example `volcengine`, `openai`, `deepseek`).
- `api_key`: Optional API key for the agent model provider. Can be configured here directly when you want bot-specific credentials.
- `api_base`: Optional API base for the agent model provider. Useful for provider gateways or custom endpoints such as VolcEngine Ark.
@@ -208,6 +209,7 @@ All configurations are under the `bot` field in `ov.conf`, with default values f
"api_key": "<your-ark-api-key>",
"api_base": "https://ark.cn-beijing.volces.com/api/v3",
"provider": "volcengine",
+ "temperature": 0.7,
"max_tool_iterations": 50,
"memory_window": 50
},
diff --git a/bot/README_CN.md b/bot/README_CN.md
index 79641870b..38d28f5e4 100644
--- a/bot/README_CN.md
+++ b/bot/README_CN.md
@@ -169,6 +169,7 @@ bot 将连接到远程 OpenViking Server,使用前请先启动 OpenViking Serv
所有配置都位于 `ov.conf` 的 `bot` 字段下,且各配置项都带有默认值。可选手动配置项说明如下:
- `agents`:Agent 配置
- `model`:bot 使用的 LLM 模型名。当设置了 `provider` 时,建议直接使用 provider 原生模型名(例如 `doubao-seed-2-0-pro-260215`)。
+ - `temperature`:LLM 请求的采样温度,默认 `0.7`。
- `provider`:可选的模型 provider 名称。设置后,vikingbot 会通过 OpenViking 的 `VLMFactory` + adapter 路径直接创建后端(例如 `volcengine`、`openai`、`deepseek`)。
- `api_key`:可选,Agent 模型 provider 的 API Key。若希望 bot 使用独立凭证,可直接在这里配置。
- `api_base`:可选,Agent 模型 provider 的 API Base。适用于 provider 网关或自定义端点,例如 VolcEngine Ark。
@@ -210,6 +211,7 @@ bot 将连接到远程 OpenViking Server,使用前请先启动 OpenViking Serv
"api_key": "<your-ark-api-key>",
"api_base": "https://ark.cn-beijing.volces.com/api/v3",
"provider": "volcengine",
+ "temperature": 0.7,
"max_tool_iterations": 50,
"memory_window": 50
},
diff --git a/bot/tests/test_agent_loop_outcome.py b/bot/tests/test_agent_loop_outcome.py
index 12f122ec6..a45d35d25 100644
--- a/bot/tests/test_agent_loop_outcome.py
+++ b/bot/tests/test_agent_loop_outcome.py
@@ -7,7 +7,7 @@ from vikingbot.agent import loop as loop_module
from vikingbot.agent.loop import AgentLoop
from vikingbot.bus.events import InboundMessage, OutboundEventType
from vikingbot.bus.queue import MessageBus
-from vikingbot.config.schema import Config, SessionKey
+from vikingbot.config.schema import AgentsConfig, Config, SessionKey
from vikingbot.heartbeat.service import HEARTBEAT_METADATA_KEY
from vikingbot.providers.base import LLMProvider, LLMResponse, ToolCallRequest
@@ -25,6 +25,19 @@ class _FakeSubagentManager:
self.kwargs = kwargs
+class _RecordingProvider(LLMProvider):
+ def __init__(self):
+ super().__init__()
+ self.calls = []
+
+ async def chat(self, *args, **kwargs):
+ self.calls.append((args, kwargs))
+ return LLMResponse(content="ok")
+
+ def get_default_model(self) -> str:
+ return "fake-model"
+
+
class _FakeLangfuseClient:
def __init__(self):
self.calls = []
@@ -73,6 +86,46 @@ class _FakeOVClient:
return {"session_id": session_id, "status": "accepted"}
+def test_agents_config_temperature_schema_caps_at_two():
+ schema = AgentsConfig.model_json_schema()
+ temperature = schema["properties"]["temperature"]
+
+ assert temperature["default"] == 0.7
+ assert temperature["minimum"] == 0.0
+ assert temperature["maximum"] == 2.0
+
+
+async def test_agent_loop_passes_configured_temperature_to_provider(temp_dir: Path, monkeypatch):
+ monkeypatch.setattr(AgentLoop, "_register_builtin_hooks", lambda self: None)
+ monkeypatch.setattr(AgentLoop, "_register_default_tools", lambda self: None)
+ monkeypatch.setattr("vikingbot.agent.loop.SubagentManager", _FakeSubagentManager)
+
+ bus = MessageBus()
+ provider = _RecordingProvider()
+ config = Config(storage_workspace=str(temp_dir), agents={"temperature": 0.2})
+ loop = AgentLoop(
+ bus=bus,
+ provider=provider,
+ workspace=temp_dir / "workspace",
+ model=config.agents.model,
+ temperature=config.agents.temperature,
+ config=config,
+ )
+
+ session_key = SessionKey(type="cli", channel_id="default", chat_id="session-1")
+ response, _, _ = await loop._chat_with_stream_events(
+ messages=[{"role": "user", "content": "hello"}],
+ tools=[],
+ session_key=session_key,
+ publish_events=False,
+ )
+
+ assert response.content == "ok"
+ assert provider.calls[0][1]["temperature"] == 0.2
+ assert loop.subagents.kwargs["temperature"] == 0.2
+
+
@pytest.mark.asyncio
async def test_agent_loop_makes_final_no_tool_call_when_iteration_limit_reached(
temp_dir: Path, monkeypatch
diff --git a/bot/vikingbot/agent/loop.py b/bot/vikingbot/agent/loop.py
index a510b6bce..7d9cbebfb 100644
--- a/bot/vikingbot/agent/loop.py
+++ b/bot/vikingbot/agent/loop.py
@@ -71,6 +71,7 @@ class AgentLoop:
provider: LLMProvider,
workspace: Path,
model: str | None = None,
+ temperature: float = 0.7,
max_iterations: int = 50,
memory_window: int = 50,
brave_api_key: str | None = None,
@@ -92,6 +93,7 @@ class AgentLoop:
provider: LLMProvider instance for making LLM calls.
workspace: Path to the workspace directory for file operations.
model: Optional model identifier. If not provided, uses the provider's default.
+ temperature: Sampling temperature for LLM requests (default: 0.7).
max_iterations: Maximum number of tool execution iterations per message (default: 50).
memory_window: Maximum number of messages to keep in session memory (default: 50).
brave_api_key: Optional API key for Brave search integration.
@@ -122,6 +124,7 @@ class AgentLoop:
self.provider = provider
self.workspace = workspace
self.model = model or provider.get_default_model()
+ self.temperature = temperature
self.max_iterations = max_iterations
self.memory_window = memory_window
self.brave_api_key = brave_api_key
@@ -146,6 +149,7 @@ class AgentLoop:
bus=bus,
config=self.config,
model=self.model,
+ temperature=self.temperature,
sandbox_manager=sandbox_manager,
)
@@ -281,6 +285,7 @@ class AgentLoop:
messages=messages,
tools=tools,
model=self.model,
+ temperature=self.temperature,
session_id=session_key.safe_name(),
):
if event.type == "content_delta":
@@ -313,6 +318,7 @@ class AgentLoop:
messages=messages,
tools=tools,
model=self.model,
+ temperature=self.temperature,
session_id=session_key.safe_name(),
)
return response, streamed_content, streamed_reasoning
@@ -1530,6 +1536,7 @@ Respond with ONLY valid JSON, no markdown fences."""
{"role": "user", "content": prompt},
],
model=self.model,
+ temperature=self.temperature,
session_id=session.key.safe_name(),
)
text = (response.content or "").strip()
diff --git a/bot/vikingbot/agent/subagent.py b/bot/vikingbot/agent/subagent.py
index ed2faf044..ca08aa13e 100644
--- a/bot/vikingbot/agent/subagent.py
+++ b/bot/vikingbot/agent/subagent.py
@@ -33,6 +33,7 @@ class SubagentManager:
bus: MessageBus,
config: "Config",
model: str | None = None,
+ temperature: float = 0.7,
sandbox_manager: "SandboxManager | None" = None,
):
from vikingbot.config.schema import ExecToolConfig
@@ -42,6 +43,7 @@ class SubagentManager:
self.bus = bus
self.config = config
self.model = model or provider.get_default_model()
+ self.temperature = temperature
self.sandbox_manager = sandbox_manager
self._running_tasks: dict[str, asyncio.Task[None]] = {}
@@ -129,6 +131,7 @@ class SubagentManager:
messages=messages,
tools=tools.get_definitions(),
model=self.model,
+ temperature=self.temperature,
)
if response.has_tool_calls:
diff --git a/bot/vikingbot/cli/commands.py b/bot/vikingbot/cli/commands.py
index 705628628..eb29ad17c 100644
--- a/bot/vikingbot/cli/commands.py
+++ b/bot/vikingbot/cli/commands.py
@@ -270,6 +270,7 @@ def _make_provider(config, langfuse_client: None = None):
p = config.agents
model = p.model if p else None
+ temperature = p.temperature if p else 0.7
api_key = p.api_key if p else None
api_base = p.api_base if p else None
provider_name = p.provider if p else None
@@ -289,6 +290,7 @@ def _make_provider(config, langfuse_client: None = None):
vlm_config: dict[str, Any] = {
"provider": provider_name,
"model": model,
+ "temperature": temperature,
}
if api_key:
vlm_config["api_key"] = api_key
@@ -443,6 +445,7 @@ def prepare_agent_loop(config, bus, session_manager, cron, quiet: bool = False,
provider=provider,
workspace=config.workspace_path,
model=config.agents.model,
+ temperature=config.agents.temperature,
max_iterations=config.agents.max_tool_iterations,
memory_window=config.agents.memory_window,
brave_api_key=config.tools.web.search.api_key or None,
diff --git a/bot/vikingbot/config/loader.py b/bot/vikingbot/config/loader.py
index 7d394c63a..82dd54f9f 100644
--- a/bot/vikingbot/config/loader.py
+++ b/bot/vikingbot/config/loader.py
@@ -131,7 +131,7 @@ def _merge_vlm_model_config(bot_data: dict, vlm_data: dict) -> None:
"""
Merge vlm model config into bot config.
- Only sets model - provider config is read directly from OpenVikingConfig.
+ Only sets model parameters - provider config is read directly from OpenVikingConfig.
"""
# Set default model from vlm.model
if "agents" in bot_data:
@@ -147,6 +147,8 @@ def _merge_vlm_model_config(bot_data: dict, vlm_data: dict) -> None:
bot_data["agents"]["provider"] = provider if provider else ""
bot_data["agents"]["api_base"] = vlm_data.get("api_base", "")
bot_data["agents"]["api_key"] = vlm_data.get("api_key", "")
+ if "temperature" in vlm_data and "temperature" not in bot_data["agents"]:
+ bot_data["agents"]["temperature"] = vlm_data["temperature"]
if "extra_headers" in vlm_data and vlm_data["extra_headers"] is not None:
bot_data["agents"]["extra_headers"] = vlm_data["extra_headers"]
diff --git a/bot/vikingbot/config/schema.py b/bot/vikingbot/config/schema.py
index d6287abd5..d09538daa 100644
--- a/bot/vikingbot/config/schema.py
+++ b/bot/vikingbot/config/schema.py
@@ -433,6 +433,12 @@ class AgentsConfig(BaseModel):
"""Agent configuration."""
model: str = "openai/doubao-seed-2-0-pro-260215"
+ temperature: float = Field(
+ default=0.7,
+ ge=0.0,
+ le=2.0,
+ description="Sampling temperature for LLM requests.",
+ )
max_tool_iterations: int = 50
memory_window: int = 50
session_context_enabled: bool = False