Component Reference

This page documents all 24 core component dataclasses used in the ECS agent architecture. All components use @dataclass(slots=True) for memory efficiency and can be imported from ecs_agent.components.

Core Agent Components

LLMComponent

Links an agent to an LLM provider for reasoning and planning.

Name	Type	Default	Description
`provider`	`LLMProvider`	(none)	The LLM provider instance
`model`	`str`	(none)	The specific model identifier
`system_prompt`	`str`	`""`	Optional system prompt override
`pending_provider`	`LLMProvider	None`	`None`
`pending_model`	`str	None`	`None`

Used by: ReasoningSystem, PlanningSystem, ReplanningSystem

Usage:

from ecs_agent.components import LLMComponent
from ecs_agent.providers.openai_provider import OpenAIProvider

llm = LLMComponent(
    provider=OpenAIProvider(api_key="..."),
    model="gpt-4o",
    system_prompt="You are a helpful assistant."
)
world.add_component(agent, llm)

# Queue model switch (takes effect at next LLM request)
llm.pending_model = "gpt-3.5-turbo"

ConversationComponent

Maintains the full history of messages for an agent's conversation.

Name	Type	Default	Description
`messages`	`list[Message]`	(none)	History of conversation messages
`max_messages`	`int`	`100`	Maximum number of messages to retain

Used by: ReasoningSystem, PlanningSystem, MemorySystem, MessageBusSystem, ToolExecutionSystem, ReplanningSystem

Usage:

from ecs_agent.components import ConversationComponent
world.add_component(agent, ConversationComponent(messages=[]))

SystemPromptComponent

Legacy component that stores the system prompt template and rendered content. For new agents, prefer SystemPromptConfigSpec with SystemPromptRenderSystem.

Name	Type	Default	Description
`template`	`str`	`""`	Base system prompt template with placeholders
`sections`	`list[PromptSectionSpec]`	`[]`	List of prompt sections to inject into placeholders
`content`	`str`	`""`	Rendered system prompt output

Used by: ReasoningSystem, PlanningSystem, ReplanningSystem (legacy path)

SystemPromptConfigSpec

New-style prompt spec using ${name} placeholder templates. Processed by SystemPromptRenderSystem.

Name	Type	Default	Description
`template_source`	`PromptTemplateSource`	required	Inline string or file path for the template
`placeholder_specs`	`dict[str, PlaceholderSpec]`	`{}`	Per-placeholder resolver specs (static, callable, or file)

Produces: RenderedSystemPromptComponent (via SystemPromptRenderSystem)

Usage:

from ecs_agent.prompts.contracts import SystemPromptConfigSpec, PromptTemplateSource

world.add_component(entity, SystemPromptConfigSpec(
    template_source=PromptTemplateSource(
        inline="You are a helpful assistant. Tools: ${_installed_tools}"
    )
))

RenderedSystemPromptComponent

cached/frozen rendered system prompt produced by SystemPromptRenderSystem on the first successful render-system pass. Reused on subsequent ticks; consumed by ReasoningSystem, PlanningSystem, ReplanningSystem.

Name	Type	Description
`text`	`str`	Fully rendered system prompt with all `${name}` placeholders resolved

RenderedUserPromptComponent

Output component written by UserPromptNormalizationSystem containing the normalized user message for the current tick. Stored conversation history is never mutated.

Name	Type	Description
`text`	`str`	Normalized user message with trigger injections prepended

`EntityRegistryComponent`

Internal registry for named entity resolution and tagging. Usually managed by World via register_entity(), not directly attached to entities.

Fields:

entity_id: EntityId — Entity being registered
name: str — Unique name for lookup
tags: set[str] — Tag set (default: empty set)
metadata: dict[str, Any] — Additional metadata (default: empty dict)

Example:

from ecs_agent.components import EntityRegistryComponent

# Usually created via world.register_entity(), not directly
agent = world.create_entity()
world.register_entity(agent, "coordinator", tags={"manager", "primary"})

Tool Components

ToolRegistryComponent

Stores the definitions and execution handlers for all tools available to the agent.

Name	Type	Default	Description
`tools`	`dict[str, ToolSchema]`	(none)	Mapping of tool names to their schemas
`handlers`	`dict[str, Callable[..., Awaitable[str]]]`	(none)	Async handlers for tool execution

Used by: ReasoningSystem, PlanningSystem, ToolExecutionSystem

Usage:

from ecs_agent.components import ToolRegistryComponent
world.add_component(agent, ToolRegistryComponent(tools=tools, handlers=handlers))

PendingToolCallsComponent

Captures tool calls generated by reasoning or planning that are waiting for execution.

Name	Type	Default	Description
`tool_calls`	`list[ToolCall]`	(none)	List of requested tool calls

Added by: ReasoningSystem, PlanningSystem
Consumed by: ToolExecutionSystem (removes the component after processing)

Usage:

from ecs_agent.components import PendingToolCallsComponent
world.add_component(agent, PendingToolCallsComponent(tool_calls=requested_calls))

ToolResultsComponent

Holds the output results of executed tool calls before they are returned to the conversation.

Name	Type	Default	Description
`results`	`dict[str, str]`	(none)	Mapping of call ID to result string

Added by: ToolExecutionSystem

Usage:

from ecs_agent.components import ToolResultsComponent
world.add_component(agent, ToolResultsComponent(results={"call_1": "Success output"}))

Planning Components

PlanComponent

Maintains a ReAct plan, tracking individual steps and completion status.

Name	Type	Default	Description
`steps`	`list[str]`	(none)	Sequential list of plan steps
`current_step`	`int`	`0`	Current active step index
`completed`	`bool`	`False`	Completion flag for the plan

Used by: PlanningSystem, ReplanningSystem

Usage:

from ecs_agent.components import PlanComponent
world.add_component(agent, PlanComponent(steps=["Analyze", "Execute", "Verify"]))

Collaboration Components

MessageBusConfigComponent

Configures the behavior and limits for the message bus.

Name	Type	Default	Description
`max_queue_size`	`int`	`1000`	Maximum number of buffered messages per subscriber
`publish_timeout`	`float`	`2.0`	Timeout in seconds for blocking publishes (backpressure)
`request_timeout`	`float`	`30.0`	Default timeout for request-response conversations
`cleanup_interval`	`float`	`10.0`	Seconds between cleanup of stale request states

Used by: MessageBusSystem

Usage:

from ecs_agent.components import MessageBusConfigComponent
world.add_component(agent, MessageBusConfigComponent(request_timeout=60.0))

MessageBusSubscriptionComponent

Tracks an entity's topic subscriptions and message queues.

Name	Type	Default	Description
`subscriptions`	`set[str]`	`set()`	Set of topics the entity is subscribed to
`inbox`	`deque[MessageBusEnvelope]`	`deque()`	Incoming message queue for the subscriber

Used by: MessageBusSystem

Usage:

from ecs_agent.components import MessageBusSubscriptionComponent
world.add_component(agent, MessageBusSubscriptionComponent(subscriptions={"agent.updates", "system.alerts"}))

MessageBusConversationComponent

Manages active request-response conversations for an entity.

Name	Type	Default	Description
`active_requests`	`dict[str, float]`	`{}`	Mapping of correlation IDs to request expiration timestamps

Used by: MessageBusSystem

Usage:

from ecs_agent.components import MessageBusConversationComponent
world.add_component(agent, MessageBusConversationComponent())

State & Lifecycle Components

ErrorComponent

Captures diagnostic information when a system encounters an exception.

Name	Type	Default	Description
`error`	`str`	(none)	The error message or traceback
`system_name`	`str`	(none)	System that triggered the error
`timestamp`	`float`	(none)	Unix timestamp of occurrence

Added by: ReasoningSystem, PlanningSystem on failure
Consumed by: ErrorHandlingSystem (processed and then removed)

Usage:

import time
from ecs_agent.components import ErrorComponent
world.add_component(agent, ErrorComponent(
    error="API key expired",
    system_name="ReasoningSystem",
    timestamp=time.time()
))

InterruptionComponent

Signals agent should stop gracefully with partial content preservation.

Name	Type	Default	Description
`reason`	`InterruptionReason`	(none)	Enum: USER_REQUESTED, SYSTEM_PAUSE, ERROR, COMPLETION
`message`	`str`	`""`	Human-readable reason string
`metadata`	`dict[str, Any]`	`{}`	Structured context (default: empty dict)
`timestamp`	`float`	(auto)	Auto-generated via time.time()

Added by: External code (user interruption), systems (error conditions) Consumed by: Runner (detects and halts execution)

Usage:

from ecs_agent.components import InterruptionComponent
from ecs_agent.types import InterruptionReason

world.add_component(agent, InterruptionComponent(
    reason=InterruptionReason.USER_REQUESTED,
    message="User clicked stop",
    metadata={"source": "web_ui"}
))

TerminalComponent

A marker component that signifies the agent has finished its task.

Name	Type	Default	Description
`reason`	`str`	(none)	Final completion status or reason

Checked by: Runner loop to determine when to stop execution.

In interactive flows, a TerminalComponent can also be handled by the opt-in TerminalCleanupSystem, which clears selected reasons after terminal-producing systems run. By default, that helper only clears reasoning_complete and leaves other terminal reasons intact.

Usage:

from ecs_agent.components import TerminalComponent
world.add_component(agent, TerminalComponent(reason="Task goal reached."))

OwnerComponent

Establishes a hierarchical relationship between entities.

Name	Type	Default	Description
`owner_id`	`EntityId`	(none)	ID of the parent/owner entity

Usage:

from ecs_agent.components import OwnerComponent
world.add_component(child, OwnerComponent(owner_id=root_agent_id))

KVStoreComponent

Provides a generic key-value store for custom system memory.

Name	Type	Default	Description
`store`	`dict[str, Any]`	(none)	Backing dictionary for storage

Used by: Available for custom implementations and extensions.

Usage:

from ecs_agent.components import KVStoreComponent
world.add_component(agent, KVStoreComponent(store={"count": 5}))

Advanced & Feature-Specific Components

ToolApprovalComponent

Configures how tool calls are approved or denied before execution.

Name	Type	Default	Description
`policy`	`ApprovalPolicy`	(none)	`ALWAYS_APPROVE`, `ALWAYS_DENY`, or `REQUIRE_APPROVAL`
`timeout`	`float	None`	`30.0`
`approved_calls`	`list[str]`	`[]`	History of approved tool call IDs
`denied_calls`	`list[str]`	`[]`	History of denied tool call IDs
,
Used by: `ToolApprovalSystem`

Usage:

from ecs_agent.components import ToolApprovalComponent
from ecs_agent.types import ApprovalPolicy

world.add_component(agent, ToolApprovalComponent(policy=ApprovalPolicy.REQUIRE_APPROVAL, timeout=60.0))

SandboxConfigComponent

Defines execution limits for code-running tools or sandboxed environments.

Name	Type	Default	Description
`timeout`	`float`	`30.0`	Maximum execution time in seconds
`max_output_size`	`int`	`10000`	Maximum allowed size of execution output in bytes
Used by: Tools using `Sandbox` logic.

Usage:

from ecs_agent.components import SandboxConfigComponent
world.add_component(agent, SandboxConfigComponent(timeout=10.0, max_output_size=5000))

PlanSearchComponent

Configures Monte Carlo Tree Search (MCTS) for finding the optimal plan path.

Name	Type	Default	Description
`max_depth`	`int`	`5`	Maximum search depth in the MCTS tree
`max_branching`	`int`	`3`	Maximum number of child nodes to expand per parent
`exploration_weight`	`float`	`1.414`	UCB1 exploration constant
`best_plan`	`list[str]`	`[]`	The best plan path found after search completes
`search_active`	`bool`	`False`	Internal flag indicating search is in progress
Used by: `TreeSearchSystem` (mutually exclusive with `PlanComponent`)

Usage:

from ecs_agent.components import PlanSearchComponent
world.add_component(agent, PlanSearchComponent(max_depth=10, max_branching=5))

RAGTriggerComponent

Triggers a vector search and stores the retrieved document snippets.

Name	Type	Default	Description
`query`	`str`	`""`	The search query string; cleared after retrieval
`top_k`	`int`	`5`	Number of documents to retrieve
`retrieved_docs`	`list[str]`	`[]`	Snippets of retrieved text
,
Used by: `RAGSystem`

Usage:

from ecs_agent.components import RAGTriggerComponent
world.add_component(agent, RAGTriggerComponent(query="How to use ECS?", top_k=3))

EmbeddingComponent

Provides an entity with access to an embedding provider. ,

Name	Type	Default	Description
`provider`	`EmbeddingProvider`	(none)	The embedding provider instance
`dimension`	`int`	`0`	Expected vector dimension
,
Used by: `RAGSystem`

Usage:

from ecs_agent.components import EmbeddingComponent
from ecs_agent.providers.fake_embedding_provider import FakeEmbeddingProvider

world.add_component(agent, EmbeddingComponent(provider=FakeEmbeddingProvider(), dimension=384))

VectorStoreComponent

Links an entity to a vector store for storage and retrieval. ,

Name	Type	Default	Description
`store`	`VectorStore`	(none)	The vector store instance
,
Used by: `RAGSystem`
,
Usage:

from ecs_agent.components import VectorStoreComponent
from ecs_agent.providers.vector_store import InMemoryVectorStore

world.add_component(agent, VectorStoreComponent(store=InMemoryVectorStore(dimension=384)))

StreamingComponent

Enables system-level streaming output for an entity's LLM responses.

Name	Type	Default	Description
`enabled`	`bool`	`False`	Whether streaming is active for this entity

Used by: ReasoningSystem (checks this to decide streaming vs batch mode)

Usage:

from ecs_agent.components import StreamingComponent
world.add_component(agent, StreamingComponent(enabled=True))

CheckpointComponent

Stores snapshots of world state for undo/restore operations.

Name	Type	Default	Description
`snapshots`	`list[dict[str, Any]]`	`[]`	Stack of serialized world state snapshots
`max_snapshots`	`int`	`10`	Maximum number of snapshots to retain

Used by: CheckpointSystem

Usage:

from ecs_agent.components import CheckpointComponent
world.add_component(agent, CheckpointComponent(max_snapshots=5))

CompactionConfigComponent

Configures when and how conversation compaction occurs.

Name	Type	Default	Description
`threshold_tokens`	`int`	(none)	Token count threshold triggering compaction
`summary_model`	`str`	(none)	Model to use for generating summaries

Used by: CompactionSystem

Usage:

from ecs_agent.components import CompactionConfigComponent
world.add_component(agent, CompactionConfigComponent(threshold_tokens=4000, summary_model="qwen-plus"))

ConversationArchiveComponent

Stores archived conversation summaries created by compaction.

Name	Type	Default	Description
`archived_summaries`	`list[str]`	`[]`	List of past conversation summaries

Used by: CompactionSystem

Usage:

from ecs_agent.components import ConversationArchiveComponent
world.add_component(agent, ConversationArchiveComponent())

RunnerStateComponent

Tracks the current state of the runner execution on an entity.

Name	Type	Default	Description
`current_tick`	`int`	(none)	Current tick number in the runner loop
`is_paused`	`bool`	`False`	Whether the runner is paused
`checkpoint_path`	`str	None`	`None`

Used by: Runner.save_checkpoint(), Runner.load_checkpoint()

Usage:

from ecs_agent.components import RunnerStateComponent
# Usually added automatically by Runner, but can be pre-configured
world.add_component(agent, RunnerStateComponent(current_tick=0))

UserInputComponent

Enables an entity to request and receive async user input.

Name	Type	Default	Description
`prompt`	`str`	`""`	The prompt to display to the user
`future`	`asyncio.Future[str]	None`	`None`
`timeout`	`float	None`	`None`
`result`	`str	None`	`None`

Used by: UserInputSystem

Usage:

from ecs_agent.components import UserInputComponent
world.add_component(agent, UserInputComponent(prompt="Enter your name: ", timeout=None))

ConversationTreeComponent

Tree-structured conversation with branching and linearization.

Name	Type	Default	Description
`messages`	`dict[str, ConversationMessage]`	`{}`	All messages indexed by ID
`current_branch_id`	`str	None`	`None`
`branches`	`dict[str, ConversationBranch]`	`{}`	All branches indexed by ID

Used by: conversation_tree module, ReasoningSystem (auto-linearizes current branch)

Usage:

from ecs_agent.components import ConversationTreeComponent
from ecs_agent.types import ConversationMessage, ConversationBranch

root_msg = ConversationMessage(
    id="msg_0",
    parent_message_id=None,
    role="user",
    content="Hello",
)
branch = ConversationBranch(branch_id="main", leaf_message_id="msg_0")

world.add_component(
    agent,
    ConversationTreeComponent(
        messages={"msg_0": root_msg},
        current_branch_id="main",
        branches={"main": branch},
    ),
)

ResponsesAPIStateComponent

Tracks OpenAI Responses API state for multi-turn conversations.

Name	Type	Default	Description
`previous_response_id`	`str	None`	`None`

Used by: OpenAIProvider (when use_responses_api=True)

Usage:

from ecs_agent.components import ResponsesAPIStateComponent

world.add_component(
    agent,
    ResponsesAPIStateComponent(previous_response_id=None),
)
# System automatically updates previous_response_id after each LLM call

SubagentRegistryComponent

Registry of named subagent configurations for delegation.

Name	Type	Default	Description
`subagents`	`dict[str, SubagentConfig]`	`{}`	Subagent configurations by name

Used by: SubagentSystem, delegate tool

Usage:

from ecs_agent.components import SubagentRegistryComponent
from ecs_agent.types import SubagentConfig

researcher = SubagentConfig(
    name="researcher",
    provider=your_provider,
    model="gpt-4o",
    system_prompt="You are a research assistant.",
    max_ticks=10,
    skills=[],
)

world.add_component(
    agent,
    SubagentRegistryComponent(subagents={"researcher": researcher}),
)

TaskComponent

The primary component for task definition and tracking in the Task Orchestration System.

Name	Type	Default	Description
`task_id`	`str`	(none)	Stable identifier for the task
`description`	`str`	(none)	What the task does (supports placeholder rendering)
`expected_output`	`str`	(none)	What success looks like
`assigned_agent`	`EntityId \| str \| None`	(none)	EntityId (local), str (subagent name), or None
`tools`	`list[str]`	(none)	List of tool names available for the task
`context_dependencies`	`list[str]`	(none)	List of scratchbook refs needed for execution
`status`	`TaskStatus`	(none)	Current state machine status
`priority`	`int`	`0`	Execution priority (higher = earlier)
`output_schema`	`dict[str, Any] \| None`	`None`	Optional Pydantic-style schema for output validation
`max_retries`	`int`	`0`	Maximum number of automatic retry attempts

Used by: TaskFetchingUnit, TaskExecutor, WavePlanner, DependencyAnalyzer

Usage:

from ecs_agent.components import TaskComponent
from ecs_agent.types import TaskStatus

task = TaskComponent(
    task_id="analyze_report",
    description="Analyze the report at {{tool_results/report_id}}",
    expected_output="Summary of the report",
    assigned_agent="analyst_agent",
    tools=["read_file", "write_file"],
    context_dependencies=["tool_results/report_id"],
    status=TaskStatus.PENDING,
    priority=10
)
world.add_component(agent, task)

ScratchbookRefComponent

Stores a reference to a specific scratchbook artifact.

Name	Type	Default	Description
`artifact_id`	`str`	(none)	Unique identifier for the artifact file
`category`	`str`	(none)	Category subfolder path
`content_hash`	`str`	(none)	SHA256 hash of the artifact content
`timestamp`	`str`	(none)	Creation timestamp

Used by: TaskPersistenceService, ContextResolver, ScratchbookIndexer

Usage:

from ecs_agent.components import ScratchbookRefComponent

ref = ScratchbookRefComponent(
    artifact_id="report_001",
    category="tool_results",
    content_hash="abc123...",
    timestamp="1709827200.0"
)
world.add_component(agent, ref)

ScratchbookIndexComponent

Maintains an index of available artifacts for an agent.

Name	Type	Default	Description
`artifacts`	`dict[str, ScratchbookRef]`	`{}`	Mapping of artifact IDs to their metadata references

Used by: ScratchbookIndexer

Usage:

from ecs_agent.components import ScratchbookIndexComponent

world.add_component(agent, ScratchbookIndexComponent(artifacts={}))

UserPromptConfigComponent

Configures opt-in prompt normalization and keyword injection.

Name	Type	Default	Description
`triggers`	`dict[str, str]`	`{}`	Mapping of `@keyword` or `event:<name>` to template content
`enable_context_pool`	`bool`	`False`	Enable one-shot context collection
`context_pool_max_chars`	`int`	`8192`	Maximum characters for context block

PromptContextQueueComponent

Queue-backed storage for context items (tool results, subagent outputs) to be injected into outbound user messages.

Name	Type	Default	Description
`entries`	`list[ContextEntry]`	`[]`	Collected context entries

PromptContextReservationComponent

Tracks an active reservation snapshot for deterministic prompt context injection.

Name	Type	Default	Description
`reservation_id`	`str`	(required)	Unique reservation identifier
`created_at_tick`	`int`	(required)	Tick when reservation was created
`reserved_entries`	`list[ContextEntry]`	(required)	Snapshot of reserved context entries

FilesExpand file tree

components.md

Latest commit

History

components.md

File metadata and controls

Component Reference

Core Agent Components

LLMComponent

ConversationComponent

SystemPromptComponent

SystemPromptConfigSpec

RenderedSystemPromptComponent

RenderedUserPromptComponent

EntityRegistryComponent

Tool Components

ToolRegistryComponent

PendingToolCallsComponent

ToolResultsComponent

Planning Components

PlanComponent

Collaboration Components

MessageBusConfigComponent

MessageBusSubscriptionComponent

MessageBusConversationComponent

State & Lifecycle Components

ErrorComponent

InterruptionComponent

TerminalComponent

OwnerComponent

KVStoreComponent

Advanced & Feature-Specific Components

ToolApprovalComponent

SandboxConfigComponent

PlanSearchComponent

RAGTriggerComponent

EmbeddingComponent

VectorStoreComponent

StreamingComponent

CheckpointComponent

CompactionConfigComponent

ConversationArchiveComponent

RunnerStateComponent

UserInputComponent

ConversationTreeComponent

ResponsesAPIStateComponent

SubagentRegistryComponent

TaskComponent

ScratchbookRefComponent

ScratchbookIndexComponent

UserPromptConfigComponent

PromptContextQueueComponent

PromptContextReservationComponent

`EntityRegistryComponent`