Report Abuse

Basic Information

Agent File (.af) is an open standard file format for serializing stateful AI agents. The repository defines the .af schema and provides example agent files designed to capture all state needed to re-create agents with persistent memory and behavior. It was created alongside the Letta framework but is aimed at portability between compatible systems. The repo includes downloadable example agents (MemGPT, Deep Research, Customer Support, Stateless Workflow, Composio Tools), import/export instructions, and guidance for adding .af support to other frameworks. It also documents what agent state is included, how secrets are handled on export, and a roadmap of planned features such as multi-agent files, archival memory passages, and framework converters.

Links

App Details

Features
The repository defines a comprehensive schema and bundles reference examples and tooling instructions. Key features documented include storage of model configuration (model name, context window, embedding model), complete message history with in-context flags, system prompt, in-context memory blocks, tool definitions including source code and JSON schemas, tool rules and sequencing, environment variables for tool execution, and handling of secrets (secrets are nulled on export). The repo provides import/export workflows via Letta ADE, cURL, Node.js (TypeScript) and Python SDK snippets, downloadable .af example files, and links to the full schema used by Letta. A roadmap lists support for MCP, passages, data sources, migrations, multi-agent files, and converters.
Use Cases
Agent File standardizes agent state so developers and teams can share, preserve, version, and migrate stateful agents across environments. By packaging prompts, memory, tools, and model settings in a single portable file, .af reduces vendor lock-in and enables collaboration and archival of agent configurations. The provided examples let users bootstrap agents or test import/export behavior. Integration examples and SDK snippets for Letta Server show practical import and export flows, enabling developers to implement .af support in their own frameworks by mapping schema fields to native representations. The roadmap and contribution guidance invite community improvements and converters to increase cross-framework compatibility.

Please fill the required fields*