Constructor
| Parameter | Type | Default | Description |
|---|---|---|---|
model | str | Model | — | LLM provider. String shorthand ("gpt-4o") or Model instance. |
instructions | str | "" | System prompt instructions. |
tools | List[Function] | [] | Tool functions decorated with @tool. |
toolkits | List[Toolkit] | [] | Toolkit instances (MCPToolkit, BrowserToolkit). |
skills | List[Skill] | [] | Skill instances (instructions + tools). |
knowledge | Knowledge | str | None | RAG knowledge base or directory path shorthand. |
memory | Memory | bool | None | Session memory. True for InMemoryStore. |
thinking | Thinking | bool | None | Reasoning phase. True for defaults. |
deep_research | DeepResearchConfig | bool | None | Multi-wave research. True for defaults. |
audio_transcriber | AudioTranscriber | bool | None | Voice transcription. True for Whisper. |
security | SecurityConfig | bool | None | Security hardening. True for defaults. |
guardrails | Guardrails | None | Input/output/tool guardrails. |
middleware | List[Middleware] | [] | Middleware chain. |
tracing | Tracing | bool | None | Event export. True for DebugExporter. |
plugins | List[Plugin] | None | Plugins loaded on first arun(). |
readers | BaseReader | bool | None | File readers. True for auto-registry. |
config | AgentConfig | AgentConfig() | Execution configuration. |
usage | UsageTracker | bool | None | Token/cost tracking. True for defaults. |
name | str | None | Display name (overrides config.agent_name). |
debug | bool | False | Color-coded turn-by-turn debug output. |
Methods
| Method | Returns | Description |
|---|---|---|
run(instruction, **kwargs) | RunOutput | Synchronous execution. |
arun(instruction, **kwargs) | RunOutput | Async execution. |
run_stream(instruction, **kwargs) | Iterator[RunOutputEvent] | Sync streaming. |
arun_stream(instruction, **kwargs) | AsyncIterator[RunOutputEvent] | Async streaming. |
serve(*interfaces, **kwargs) | — | Start runtime with interfaces, webhooks, cron. |
aserve(*interfaces, **kwargs) | — | Async version of serve. |
replay(run_output=, trace_file=) | Replay | RunOutput | Inspect or re-execute a past run. |
compare(a, b) | ReplayComparison | Compare two runs. |
security_audit() | SecurityReport | Run security audit. |
use(middleware) | Agent | Add middleware (fluent). |
Run Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
instruction | str | — | User message. |
messages | List[Message] | None | Existing conversation history. |
session_id | str | None | Session grouping identifier. |
run_id | str | None | Unique run identifier. |
user_id | str | None | User identifier for memory scoping. |
images | List[Image] | None | Image inputs. |
audio | List[Audio] | None | Audio inputs. |
videos | List[Video] | None | Video inputs. |
files | List[File] | None | File inputs. |
output_schema | Type[BaseModel] | None | Pydantic model for structured output. |
Properties
| Property | Type | Description |
|---|---|---|
agent_id | str | Unique identifier. |
agent_name | str | Display name. |
tool_names | List[str] | Registered tool names. |
plugin_registry | PluginRegistry | Plugin manager. |
usage_tracker | UsageTracker | None | Token/cost tracker. |
scheduler | Scheduler | None | Auto-created scheduler for triggers. |