AI Agent Memory
AI agent memory is what lets an agent remember the current conversation, recall relevant facts it learned before, and keep working notes across a long task. agentFast ships memory as tiers, because a conversation buffer and a long-term knowledge store are genuinely different problems.
Short-term buffer
The recent conversation, session-scoped — the working context of the current exchange.
Long-term facts
User-scoped facts for semantic recall, injected with a visible memory:recall trace step so you always see what was remembered and why.
Episodic log
An append-only record of what happened across runs, for audit and for the agent to reason over its own history.
Scratchpad tier
A file-backed working area for long autonomous runs that need to park intermediate artifacts.
pgvector or Chroma
Long-term recall runs on pgvector in your Postgres by default, with a Chroma backend by config.
Graceful FTS fallback
With no embedder configured, retrieval degrades to Postgres full-text search — nothing hard-requires an embedding provider.
Two compaction modes
Tiered for conversations, reset-artifact for long autonomous runs — wired into both the LangGraph and vanilla adapters.
kb/ and they're ingested for grounded recall — the same path the SupportAgent uses to answer from your policies.Memory pairs with observability (recalls are visible steps) and works on every SDK, including LangGraph and CrewAI.
agentFast is the production layer — memory, 50 tools, observability, evals, guardrails, human-in-the-loop — for LangGraph, CrewAI, the Claude & OpenAI Agent SDKs & Vanilla. Own it for life.
Get agentFast — from $89