ant_ai.agent.base
SubagentTask
pydantic-model
Bases: BaseModel
Show JSON schema:
{
"properties": {
"message": {
"description": "The complete task. The subagent sees nothing else of the conversation, so include every detail it needs.",
"title": "Message",
"type": "string"
}
},
"required": [
"message"
],
"title": "SubagentTask",
"type": "object"
}
Fields:
-
message(str)
Source code in src/ant_ai/agent/base.py
46 47 48 49 50 | |
message
pydantic-field
message: str
The complete task. The subagent sees nothing else of the conversation, so include every detail it needs.
BaseAgent
pydantic-model
Bases: BaseModel
Base class for all agent implementations.
Owns the public invocation and streaming APIs, hook chain, and
identity fields. Subclasses implement _make_loop() to provide the
reasoning loop.
Config:
arbitrary_types_allowed:True
Fields:
-
name(str) -
system_prompt(str) -
llm(ChatLLM) -
description(str) -
tools(list[Tool | BaseAgent]) -
hooks(list[AgentHook]) -
memory(Memory | None) -
planning(bool) -
skills(Path | list[Path] | None) -
skill_presenter(SkillPresenter) -
max_retries(int)
Validators:
-
_agents_as_tools→tools -
_build
Source code in src/ant_ai/agent/base.py
53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 | |
name
pydantic-field
name: str = 'BaseAgent'
Display name used in routing and observability.
system_prompt
pydantic-field
system_prompt: str
System instruction for the LLM.
description
pydantic-field
description: str = 'A base agent.'
Human-readable description of the agent for documentation.
tools
pydantic-field
Tools available to the agent. An agent in the list becomes its as_tool() with the default settings.
memory
pydantic-field
memory: Memory | None = None
Long-term memory kept across conversations; its tools are added to the agent's (see Memory).
planning
pydantic-field
planning: bool = False
Let the agent keep a plan of its task, saved in .plans/<agent>/<session_id>.md.
skills
pydantic-field
skills: Path | list[Path] | None = None
Path (or list of paths) to skills directories (agentskills.io standard).
skill_presenter
pydantic-field
skill_presenter: SkillPresenter
Controls how skills are injected into the system prompt.
max_retries
pydantic-field
max_retries: int = 3
Maximum number of times to retry after a hook returns RETRY.
system_message
property
system_message: Message
Return this agent's system message, with skill discovery appended if present.
stream
async
stream(
state: State,
*,
max_steps: int = 10,
ctx: InvocationContext | None = None,
response_schema: type[BaseModel] | None = None,
) -> AsyncGenerator[BaseEvent]
Stream events for a single agent turn.
Live ContentDeltaEvents are interleaved before the terminal
ReasoningEvent/ToolCallingEvent/FinalAnswerEvent they build up
to. Three independent things can force the whole-event path instead,
each for its own reason:
- A registered hook needs the complete response to decide (see
HookLayer.is_stream_safe()/AgentHook.stream_safe) — tokens already streamed to a client cannot be retracted. - Structured output's final coercion step never streams, since it may silently rewrite the raw text after the fact.
- Over A2A,
ContentDeltaEvents are only forwarded to the wire when the executor was built withstream_artifacts=True— this method still emits them either way.
Pass response_schema to request structured output on the
FinalAnswerEvent. When provided, the final event content will be
valid JSON matching that schema.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
state
|
State
|
Current agent state, including message history and other execution context. |
required |
max_steps
|
int
|
Maximum number of loop steps before stopping. |
10
|
ctx
|
InvocationContext | None
|
Invocation context for the current run. |
None
|
response_schema
|
type[BaseModel] | None
|
Schema for structured output on the
|
None
|
Yields:
| Name | Type | Description |
|---|---|---|
Event |
AsyncGenerator[BaseEvent]
|
Events produced during execution, including intermediate |
AsyncGenerator[BaseEvent]
|
updates and the final answer event. Each event's |
|
AsyncGenerator[BaseEvent]
|
is set to this agent's name, and its |
|
AsyncGenerator[BaseEvent]
|
|
Source code in src/ant_ai/agent/base.py
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 | |
ainvoke
async
ainvoke(
state: State,
*,
max_steps: int = 10,
ctx: InvocationContext | None = None,
response_schema: type[BaseModel] | None = None,
) -> str
Run a single agent turn and return the final answer.
This is the asynchronous counterpart to invoke. Internally, it
consumes stream() and returns the content of the FINAL_ANSWER
event.
Pass response_schema to request structured output on the
FINAL_ANSWER event. When provided, the returned string will contain
valid JSON matching that schema.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
state
|
State
|
Current agent state, including message history and other execution context. |
required |
max_steps
|
int
|
Maximum number of loop steps before stopping. |
10
|
ctx
|
InvocationContext | None
|
Invocation context for the current run. |
None
|
response_schema
|
type[BaseModel] | None
|
Schema for structured output on the
|
None
|
Returns:
| Type | Description |
|---|---|
str
|
The content of the |
Source code in src/ant_ai/agent/base.py
216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 | |
invoke
invoke(
state: State,
*,
max_steps: int = 10,
ctx: InvocationContext | None = None,
response_schema: type[BaseModel] | None = None,
) -> str
Run a single agent turn and return the final answer.
This is the synchronous counterpart to ainvoke. Internally, it runs
the asynchronous invocation path and returns the content of the
FinalAnswerEvent.
Pass response_schema to request structured output on the
FinalAnswerEvent. When provided, the returned string will contain
valid JSON matching that schema.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
state
|
State
|
Current agent state, including message history and other execution context. |
required |
max_steps
|
int
|
Maximum number of loop steps before stopping. |
10
|
ctx
|
InvocationContext | None
|
Invocation context for the current run. |
None
|
response_schema
|
type[BaseModel] | None
|
Schema for structured output on the
|
None
|
Returns:
| Type | Description |
|---|---|
str
|
The content of the |
Raises:
| Type | Description |
|---|---|
RuntimeError
|
If called from a thread with an active event loop. |
Source code in src/ant_ai/agent/base.py
258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 | |
as_tool
as_tool(
*,
name: str | None = None,
description: str | None = None,
max_steps: int = 10,
) -> Tool
Return this agent as a tool of another agent.
Each call runs the agent on a new conversation that holds only the
task, and returns only its final answer: the files it reads and the
tools it calls stay out of the caller's conversation. The caller's
InvocationContext is passed on, so the run shares the session and
the trace. An agent put in the tools of another agent becomes this
tool with the default settings.
Example:
explorer = Agent(name="explorer", description="Searches the code base.",
system_prompt="...", llm=llm, tools=[FilesystemTool()])
lead = Agent(name="lead", system_prompt="...", llm=llm,
tools=[explorer]) # or explorer.as_tool(max_steps=20)
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
name
|
str | None
|
Tool name, by default the agent's name. Providers accept only
letters, digits, |
None
|
description
|
str | None
|
Tool description, by default the agent's description. |
None
|
max_steps
|
int
|
Step budget of each run. |
10
|
Returns:
| Type | Description |
|---|---|
Tool
|
A tool with one argument, |
Raises:
| Type | Description |
|---|---|
ValueError
|
The agent was built with |
Source code in src/ant_ai/agent/base.py
308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 | |
add_tool
abstractmethod
Register a tool with the agent; an agent is added as its as_tool().
Source code in src/ant_ai/agent/base.py
382 383 384 385 | |