ZeptoClaw is an ultra-lightweight personal AI assistant designed to run as a single binary of approximately 6MB. Targeted at builders, self-hosters, security-conscious teams, and multi-tenant operators, it offers a complete AI agent runtime with a focus on minimal resource usage, security, and extensibility. ZeptoClaw is written in Rust and features over 3,100 tests, promising memory safety and efficient performance even under heavy load.
The platform supports integration with nine large language model providers, including Claude (Anthropic) and OpenAI, and features automatic retry and fallback between providers with exponential backoff, token budget tracking, and cost estimation. Users can interact with ZeptoClaw through nine communication channels, such as Telegram, Slack, Discord, WhatsApp, Lark, Email, Webhook, and the command line interface. Real-time SSE streaming is available for responses from Claude and OpenAI, allowing users to see token-by-token output in various modes, including CLI, gateway, and batch processing.
ZeptoClaw enables the delegation of subtasks to specialized sub-agents through agent swarms, supporting role-specific prompts and tool whitelists, with recursion blocking to prevent infinite loops. Its plugin system allows extensions via JSON manifest plugins, which are auto-discovered at startup and can define custom tools with command templates and parameter schemas. The tool comes with four built-in agent templates (coder, researcher, writer, analyst) and supports custom JSON templates, giving users control over system prompts, models, tokens, and temperature settings.
Security features include policy-based tool gating, SSRF prevention, path traversal detection, and shell command blocklists. Optional container isolation is available via Docker or Apple Container for full sandboxing. ZeptoClaw also offers persistent long-term memory with categories and tags, conversation history with session discovery and search, and batch processing for handling hundreds of prompts from text or JSONL files. Telemetry is supported through Prometheus and JSON metrics export, along with per-model cost estimation and token budget enforcement.
0. ZeptoClaw aims to combine the integrations, security, and minimalism of previous AI assistants without their tradeoffs.
In the Autonomous agents & workflows space, ZeptoClaw takes a focused approach. It focuses on running advanced AI agent workflows locally without large dependencies or cloud reliance. ZeptoClaw is an open-source project aimed at developers. The project is open source (Apache-2.0). ZeptoClaw is available on the command line, and it can be self-hosted.
ZeptoClaw first shipped in 2026. Development happens publicly on GitHub with 641 stars and 156 commits in the last 90 days. Among its 9 catalogued features are Multi-provider LLM, streaming responses, and agent swarms.
Summary written by a language model from the project’s public pages.
What PulseGate has recorded for this listing
Closest matches by what these projects do