Persistent agent memory 3

Versioned filesystem for persistent, collaborative AI agent workspaces Mesa is a versioned filesystem and virtual filesystem platform built for AI agents. It allows agents to access persistent files, create isolated branches, run parallel agent swarms, track changes, request human approvals, and roll back work through one API. Mesa supports POSIX-compatible filesystem access, large files, shared memories, documents, code, datasets, models, and media. It provides sub-50ms reads and writes through MesaFS, instant workspace forks, strong consistency, fine-grained access controls, and enterprise deployment options.

AI-native memory platform for stateful agents and continual context Honcho is an AI-native memory platform for building stateful AI agents. It stores and indexes messages, continuously reasons about interactions through its Neuromancer models, and returns curated context through a simple context() method. Honcho helps agents maintain continuity, understand user preferences, reduce unnecessary token usage, and support advanced reasoning across users, agents, NPCs, groups, and other entities. It supports model-agnostic integrations, unlimited retrieval, token budgets, peer scoping, background dreaming, and on-demand reasoning.

Platform for building and deploying internal AI agents Keystroke is an all-in-one platform for building, deploying, and managing AI agents and internal AI systems. Users can describe the agent they need or build one in plain TypeScript, connect it to tools and services, add memory, files, workflows, triggers, approvals, and integrations, then test and deploy it to a shared workspace. Keystroke supports more than 1,000 integrations, APIs, and MCP servers, along with multi-agent orchestration, persistent workspaces, browser use, web search, external messaging channels, usage controls, audit trails, and enterprise governance.