Crate traitclaw 

TraitClaw

A Rust AI Agent Framework — Simple by default, powerful when needed.

This is the main entry point for the TraitClaw framework. It re-exports everything from traitclaw-core and traitclaw-macros, so you only need one dependency in your Cargo.toml:

[dependencies]
traitclaw = "1.0"

Feature Flags

Feature	Crate	Default
openai-compat	traitclaw-openai-compat	✅
macros	traitclaw-macros	✅
steering	traitclaw-steering	❌
sqlite	traitclaw-memory-sqlite	❌
mcp	traitclaw-mcp	❌
rag	traitclaw-rag	❌
team	traitclaw-team	❌
eval	traitclaw-eval	❌
strategies	traitclaw-strategies	❌
full	all of the above	❌

Re-exports

pub use traitclaw_openai_compat as openai_compat;

pub use traitclaw_steering as steering;

pub use traitclaw_memory_sqlite as memory_sqlite;

pub use traitclaw_mcp as mcp;

pub use traitclaw_rag as rag;

pub use traitclaw_team as team;

pub use traitclaw_eval as eval;

pub use traitclaw_strategies as strategies;

Modules

agent

Agent — the main entry point for running AI agents.

agent_builder

Agent builder for fluent configuration.

config

Agent configuration.

context_managers

Built-in ContextManager implementations for common compression strategies.

default_strategy

Default strategy — preserves v0.1.0 agent loop behavior.

error

Error types for TraitClaw Core.

factory

Agent factory for shared-provider multi-agent creation.

memory

Memory implementations.

pool

Agent pool for managing and executing groups of agents.

prelude

Prelude module for convenient imports.

registries

Built-in ToolRegistry implementations.

retry

Retry wrapper for LLM providers with exponential backoff.

token_counter

Accurate tiktoken-based token counting.

token_counting

Token counting utilities for context management.

traits

Core trait definitions for TraitClaw.

transformers

Built-in OutputTransformer implementations for common use cases.

types

Shared types for TraitClaw.

Structs

AdaptiveRegistry

A tool registry that automatically limits the number of visible tools based on the configured model tier.

AdaptiveStrategy

Adaptive strategy that queries Tracker::recommended_concurrency() to decide whether to run sequentially or in parallel.

Agent

The main agent struct.

AgentBuilder

Builder for constructing an Agent with a fluent API.

AgentFactory

A factory for creating multiple agents from a shared provider.

AgentOutput

Output from an agent run.

AgentPool

A collection of agents for group execution.

AgentRuntime

Runtime context provided to strategies.

AgentSession

A session-scoped agent wrapper.

AgentState

Snapshot of the agent’s runtime state.

BudgetAwareTruncator

Truncates output to a maximum character count, respecting context utilization.

CharApproxCounter

Approximate token counter using the 4-characters ≈ 1-token heuristic.

CompletionRequest

A completion request to send to an LLM provider.

CompletionResponse

A completion response from the LLM provider.

CompressedMemory

A memory decorator that compresses older messages into summaries.

DefaultStrategy

The default agent strategy preserving v0.1.0 behavior.

DynamicRegistry

A mutable tool registry supporting runtime add/remove/toggle.

FullOutputRetriever

Virtual tool that retrieves the full output cached by ProgressiveTransformer.

GroupedRegistry

A tool registry that organizes tools into named groups with group-level activation and deactivation.

HintMessage

A hint message to be injected into the conversation.

InMemoryMemory

In-memory implementation of the Memory trait.

JsonExtractor

Extracts JSON from tool output, discarding surrounding text.

LlmCompressor

Summarizes old messages using an LLM provider.

LoggingHook

A hook that logs all lifecycle events using tracing.

Message

A single message in a conversation.

ModelInfo

Information about the LLM model being used by a provider.

ParallelStrategy

Execute tool calls concurrently with bounded concurrency.

PendingToolCall

A pending tool call to be executed by a strategy.

ProgressiveTransformer

A two-phase output transformer that returns an LLM-generated summary first, with the full output cached and available on demand via the __get_full_output virtual tool.

RetryConfig

Configuration for retry behavior.

RetryProvider

A provider decorator that retries transient errors with exponential backoff.

RuleBasedCompressor

Scores messages by importance and removes lowest-scored first.

RunUsage

Usage statistics from an agent run.

SequentialStrategy

Execute tool calls one at a time in order.

SimpleRegistry

A simple, immutable tool registry that wraps Vec<Arc<dyn ErasedTool>>.

TierLimits

Per-tier tool limits.

TieredCompressor

Chains compression tiers: keep recent → rule-compress mid → LLM-summarize old.

TikTokenCounter

Exact token counter using OpenAI-compatible BPE tokenization via tiktoken-rs.

ToolCall

A tool call requested by the LLM.

ToolResult

The result of executing a single tool call.

ToolSchema

JSON Schema representation for a tool’s parameters.

TransformerChain

Pipes output through multiple transformers in order.

Usage

Token usage statistics for a completion.

Enums

Action

An action that the agent wants to perform.

AgentOutputContent

The content type of an agent output.

Error

Errors that can occur during agent operation.

GuardResult

Result of a guard check.

HintPriority

Priority level for hint injection.

HookAction

The result of a hook’s interception decision.

InjectionPoint

Where in the conversation to inject the hint message.

MessageRole

Role of a message in a conversation.

ModelTier

Model capability tier used by the steering system to auto-configure Guard, Hint, and Tracker behavior.

StreamEvent

Events emitted during streaming.

Traits

AgentHook

Async lifecycle hooks for agent observability and interception.

AgentStrategy

Pluggable agent execution strategy.

ContextManager

Async trait for pluggable context window management.

ErasedTool

Type-erased tool wrapper for dynamic dispatch.

ExecutionStrategy

Trait for pluggable tool execution strategies.

Guard

Hard boundary check for agent actions.

Hint

Contextual guidance injection for model steering.

Memory

Trait for the 3-layer memory system.

OutputTransformer

Async trait for context-aware tool output transformation.

Provider

Trait for LLM providers.

TokenCounter

Trait for pluggable token counting backends.

Tool

Trait for defining agent tools with typed inputs and outputs.

ToolRegistry

Trait for pluggable tool management.

Tracker

Runtime state monitor for the steering system.

Type Aliases

CompletionStream

A stream of completion events.

Result

Convenience type alias for Result<T, Error>.

Derive Macros

Tool

Derive the ErasedTool implementation boilerplate for a struct.