Expand description
Rust-flavored building blocks for running AGNO-style agents.
The crate provides a minimal runtime with:
- A language model abstraction (
LanguageModel). - A simple tool interface (
ToolandToolRegistry). - An
Agentthat loops between the model and tools using structured JSON directives.
Modules§
- guardrails
- Guardrails module for input/output validation.
- mcp
- MCP (Model Context Protocol) client support for sayr-engine.
- reasoning
- Reasoning module for chain-of-thought agent orchestration.
- tools
- Tools module - provides various toolkits for agents.
Structs§
- Access
Controller - Agent
- An AGNO-style agent that alternates between the LLM and registered tools.
- Agent
Runtime - Agent
Task - Task that dispatches to an individual agent and stores the reply under a key.
- AppConfig
- Attachment
- A non-textual payload that can accompany a message.
- AwsBedrock
Client - AWS Bedrock client. Currently optimized for Anthropic Claude 3 models on Bedrock.
- Azure
OpenAI Client - Azure OpenAI client for Azure-hosted models.
- Cohere
Client - Conversation
Memory - In-memory transcript storage.
- Deployment
Config - Deployment
Plan - Document
- Evaluation
Report - Fallback
Chain - File
Conversation Store - A simple JSONL-based store that writes messages to disk.
- Fireworks
Client - Fireworks AI client using their OpenAI-compatible API. Default model: accounts/fireworks/models/llama-v3p1-70b-instruct
- Full
Memory Strategy - Keep all messages (default, no limiting)
- Function
Task - Wrap a plain async function as a workflow task.
- Groq
Client - Groq client - uses OpenAI-compatible API with Groq’s endpoint. Default model: llama-3.3-70b-versatile
- InMemory
Vector Store - Knowledge
Base - Message
- A single message in the conversation transcript.
- Metrics
Tracker - Mistral
Client - Mistral AI client using their OpenAI-compatible API. Default model: mistral-large-latest
- Model
Completion - Result of a chat completion request.
- Model
Config - Ollama
Client - Ollama client for local LLM inference. Default model: llama3.1
- OpenAI
Client - Open
AiEmbedder - Embedder that delegates to an OpenAI-compatible embedding client.
- Persistent
Conversation Memory - A conversation memory that persists messages through a pluggable backend.
- PgVector
Store - Adapter for Postgres/pgvector style databases.
- Principal
- Privacy
Rule - Provider
Config - Qdrant
Store - Adapter for Qdrant (or other HTTP/gRPC vector databases).
- Retrieval
Config - Retrieval
Evaluation - Retrieval
Overrides - Retry
Policy - Scored
Document - Search
Params - Security
Config - Server
Config - Sliding
Window Chunker - Token (word) based chunker with sliding window overlap.
- SqlConversation
Store - Placeholder for SQL-based backends. The type compiles without requiring the database drivers and can be swapped out once the feature lands.
- Stub
Model - Summarized
Memory Strategy - Keep first and last N messages, summarize the middle
- Team
- A coordination surface for multiple agents that share context and a message bus.
- Telemetry
Collector - Telemetry
Config - Telemetry
Labels - Telemetry
Sink - Together
Client - Together AI client using their OpenAI-compatible API. Default model: meta-llama/Llama-3.3-70B-Instruct-Turbo
- Token
Limited Memory Strategy - Token-based memory limiting (approximate)
- Tool
Call - A tool call generated by the language model.
- Tool
Description - Static description of a tool that can be embedded in prompts.
- Tool
Registry - Tool
Result - A tool result message captured in the transcript.
- Transformer
Embedder - Embedder that wraps a transformer runtime (e.g., candle, ort, ggml).
- Whitespace
Embedder - Basic whitespace tokenizer with hashed buckets for deterministic embeddings.
- Windowed
Memory Strategy - Keep only the last N messages (sliding window)
- Workflow
- Workflow
Context - Shared state threaded through a workflow execution.
Enums§
- Action
- Agent
Directive - Structured instructions the language model should emit.
- Agno
Error - Attachment
Kind - Types of attachments supported by the runtime.
- Governance
Role - Role
- Chat roles supported by the runtime.
- Similarity
Metric - Team
Event - Events emitted by the team bus.
- Workflow
Node
Traits§
- Agent
Hook - Confirmation
Handler - Conversation
Store - Generic persistence contract for conversation state.
- Document
Chunker - Embedder
- Language
Model - Minimal abstraction around a chat completion provider.
- Memory
Strategy - Memory strategy trait for managing conversation context
- Open
AiEmbedding Client - PgVector
Client - Qdrant
Client - Retriever
- Tool
- Transformer
Client - Vector
Store - Workflow
Task