FFAI Documentation

FFAI is a declarative multi-provider AI client with named-prompt context assembly, async DAG execution, RAG, agentic tool-call loops, and built-in cost/usage tracking.

Getting Started

• Quick Start
  • Named prompt references
  • Multi-step with dependencies
  • Structured output
  • Fallback models
  • Next steps

Tutorials

• Building a RAG Pipeline with FFAI
  • Prerequisites
  • Complete listing
  • Next steps
• Memory Vector Recall: Remember What Was Discussed
  • Prerequisites
• Building a DAG Execution Pipeline with FFAI
  • Prerequisites
  • Complete listing
  • Next steps
• Building an Agent with Tool Calling
  • Prerequisites
  • Complete listing
  • Next steps
• Getting Structured Output from LLMs
  • Prerequisites
  • Complete listing
  • Next steps
• Async Patterns with FFAI
  • Prerequisites
  • Complete listing
  • Next steps
• Conditional Execution with the Condition DSL
  • Prerequisites
  • Complete listing
  • Next steps

Guides

• Installation
  • Optional extras
  • Vector store backends
  • Running tests
  • Requirements
• Configuration
  • Resolution order
  • Config file discovery
  • Config files
  • Retry settings
  • Client configuration
  • Paths configuration
  • Model defaults
  • Logging configuration
  • Observability configuration
  • RAG configuration
  • API key environment variables
  • See also
• Memory Vector Recall
  • How it works
  • Configuration
  • Embedding backend resolution
  • Persistence
  • TurnHit reference
  • Memory class API
  • Persistence API
  • Lifecycle and resource cleanup
  • Troubleshooting
  • See also
• Vector Store Backends
  • Available backends
  • Choosing a backend
  • Using get_store()
  • Configuration via YAML
  • Backend-specific notes
  • VectorStoreBase interface
  • Docker setup for pgvector and Qdrant
  • Example notebooks
  • See also
• RAG Search Strategies
  • Vector search (default)
  • BM25 keyword search
  • Hybrid search with bm25_alpha
  • Reciprocal rank fusion
  • Rerankers
  • Query expansion
  • Fuse results from multiple searches
  • See also
• Chunking Strategies
  • Choosing a strategy
  • Getting a chunker
  • Recursive chunker (default)
  • Character chunker
  • Markdown chunker
  • Code chunker
  • Hierarchical chunker
  • Using with RAG
  • Tuning chunk_size and chunk_overlap
  • See also
• Fallback Models & Client Switching
  • Fallback models
  • Per-call model override
  • Runtime client switching
  • See also
• Cost & Usage Tracking
  • Per-call tracking
  • Aggregated stats
  • DataFrame export
  • Persistence
  • RAG query tracking
  • See also
• History & Query API
  • Query methods
  • DataFrame export
  • Persistence
  • Client conversation history
• Response Validation
  • Basic validation
  • Re-execution on failure
  • Validation with retries
  • Writing good criteria
  • See also

API Reference

• API Reference
  • Clients
  • Clients.AsyncFFLiteLLMClient
  • Clients.BaseLiteLLMClient
  • Clients.FFLiteLLMClient
  • Clients.FFMistralSmall
  • Clients.model_defaults
  • ConversationHistory
  • FFAI
  • FFAIClientBase
  • OrderedPromptHistory
  • agent
  • agent.agent_loop
  • agent.agent_result
  • agent.response_validator
  • config
  • core
  • core._async
  • core.async_client_base
  • core.async_executor
  • core.client_base
  • core.condition_evaluator
  • core.conversation_manager
  • core.embeddings
  • core.execution_result
  • core.execution_state
  • core.graph
  • core.graph_execution_helpers
  • core.history
  • core.history.conversation
  • core.history.ordered
  • core.history.permanent
  • core.history.recorder
  • core.history_exporter
  • core.memory
  • core.memory.memory
  • core.memory.persist
  • core.memory.turn_store
  • core.memory.types
  • core.prompt_builder
  • core.prompt_node
  • core.prompt_utils
  • core.response_context
  • core.response_executor
  • core.response_options
  • core.response_result
  • core.response_utils
  • core.structured_output
  • core.types
  • core.usage
  • observability
  • observability.log_context
  • observability.telemetry
  • rag
  • rag.client_adapter
  • rag.embed
  • rag.format
  • rag.indexing
  • rag.indexing.bm25
  • rag.indexing.contextual
  • rag.indexing.deduplication
  • rag.indexing.hierarchical
  • rag.prompts
  • rag.rag
  • rag.search
  • rag.search.hybrid
  • rag.search.query_expansion
  • rag.search.rerankers
  • rag.splitters
  • rag.splitters.base
  • rag.splitters.character
  • rag.splitters.code
  • rag.splitters.factory
  • rag.splitters.hierarchical
  • rag.splitters.markdown
  • rag.splitters.recursive
  • rag.store
  • rag.types
  • retry_utils
  • tools
  • tools.tool_registry

Indices and tables

• Index

• Module Index

• Search Page