ESP-Claw Edge Agent Framework
Define device behavior through chat and run sensing, decisions, and actions on Espressif chips.
- Source repo
- espressif/esp-claw
- Stars
- ★ 2.3k
- Last updated
- 4d ago
- License
- Apache-2.0
- Primary language
- C
- FA score
- 37/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesOpenAI API · Claude API
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- IoT hobbyists who want to define device behavior through messaging instead of writing a program.
- Not a fit if
- Projects that need to run directly on non-Espressif chips
- Teams that need to use the agent without configuring a model API
- Source review
- 37/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
ESP-Claw is Espressif's C-based Chat Coding agent framework for IoT devices, bringing an Agent Runtime onto Espressif chips. Users can define device behavior through instant messaging, with support for dynamic Lua loading. Its event-driven Agent Loop can be triggered by events, with responses described as fast as milliseconds in some cases. The framework supports structured memory and says that memory stays off the cloud. Devices can communicate over MCP in either server or client mode. ESP-Claw targets ESP32-series development boards, with browser-based flashing for supported boards or a local firmware build path.
ESP-Claw runs an Agent Runtime on Espressif chips and responds to events that can trigger its Agent Loop. Users can chat with a device through Telegram, QQ, Feishu, or WeChat and use dynamic Lua loading to define device behavior. The framework supports OpenAI-style and Anthropic-style APIs, as well as custom endpoints; it can connect to MCP-compatible devices as either an MCP Server or Client. The resulting behavior executes on the device. Supported boards can be configured and flashed in a browser, while firmware can also be built and flashed locally.
- IoT hobbyists who want to define device behavior through messaging instead of writing a program.
- Embedded developers who need ESP32-series devices to react to sensor or other events.
- IoT prototyping teams that want local decision logic on Espressif chips.
- Developers connecting ESP32 devices to other MCP-compatible devices.
- Makers who want to configure and flash a supported board in a browser.
How do you install or deploy this agent?
The README does not provide a copyable install command, complete local build commands, model API credential steps, or a first-run command. It describes two deployment paths: choose a supported board in application/edge_agent/boards/ and configure and flash it through the browser flashing page, or follow the local build documentation to adapt a board, build, and flash. The device must use an ESP32-series chip. Model features also require an OpenAI-style or Anthropic-style API, but the supplied material does not explain credential setup.
How do you use this agent?
The README gives no copyable first-chat, API-call, or configuration example. It says users can interact with the device through Telegram, QQ, Feishu, or WeChat and define behavior in conversation; the framework supports dynamic Lua loading and an event-driven Agent Loop. To use it, flash the firmware, configure a model API endpoint, and connect the chosen messaging platform. The supplied material does not include the required configuration fields or commands.
What are this agent's strengths and limitations?
- The Agent Runtime runs on Espressif chips, allowing sensing, decisions, and actions to take place at the device.
- Dynamic Lua loading lets users define device behavior through messaging conversations.
- The event-driven Agent Loop is described as capable of millisecond responses.
- Supports OpenAI-style and Anthropic-style APIs, custom endpoints, and both MCP Server and Client modes.
- Supported boards can be configured and flashed in a browser without compiling firmware locally first.
- The core runtime targets Espressif chips; moving it directly to other chips requires adaptation.
- Model capabilities depend on an external API or custom endpoint; the README does not state costs or describe an offline model path.
- Specific commands for local builds, model credentials, messaging platform connections, and first run are not included in the supplied material.
- The project is under active development, and the README gives no stability commitment.
How does this agent compare with similar options?
The README contrasts ESP-Claw with traditional IoT devices that passively execute commands and says it was inspired by OpenClaw, with parts of the implementation drawing on MimiClaw. It does not provide feature or performance comparisons with those projects.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| ESP-Claw Edge Agent Framework This agent | 37 · Major gaps | Framework | ★ 2.3k | 4d ago | C | OpenAI API · Claude API |
| Grok CLI | 52 · Major gaps | CLIFree + model costs | ★ 3.5k | 5mo ago | TypeScript | — |
| Franklin Agent | 43 · Major gaps | CLIFreemium | ★ 561 | today | TypeScript | — |
| CommitLore | 89 · Good | CLIFree | ★ 18 | today | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes local device execution, structured memory kept off the cloud, chat, MCP, and extensible modules. It does not explain the permission model, sensitive credential storage, data transmission boundaries, or confirmation before external actions, so these controls receive only partial credit. The supplied files show no rollback mechanism. The README attributes inspiration to OpenClaw and implementation influence to MimiClaw, but gives no more detailed source mapping.
The README names supported chip families, API styles, models, and messaging platforms, and says self-programming depends on models with strong tool use. However, performance claims such as millisecond response are unsupported in the supplied material, and there is no assessable failure-message strategy or dependency availability guarantee.
The materials identify ESP32 devices, ordinary users configuring behavior through chat, and local builders as target contexts; they also mention Board Manager, online flashing, and component trimming. Agent capability boundaries, event trigger conditions, and handling of errors or risky actions are not described, so the stronger scores are limited to audience and environment fit.
The README organizes material by features, quick start, platforms, development plan, and security, and points to online flashing and local build documentation. Examples and FAQs are only linked through external documentation, known limitations are sparse, naming is consistent but no version change history is shown. The LICENSE file supplies the full Apache-2.0 terms. Issues and a Chinese survey are offered as maintenance channels, but no responsible team or maintenance commitment is stated.
The project describes a plausible IoT use for combining conversational configuration, event-driven behavior, MCP, and component extensions on low-cost ESP32 devices. Reusable outputs, demonstrated results, and quantified cost-benefit evidence are absent, which limits the scores.
Some claims are traceable to README feature descriptions, platform lists, and quick-start guidance, and the LICENSE file supports the license claim. The supplied materials do not provide independent implementation, test, or result evidence to corroborate core capabilities; some performance and privacy statements remain unsupported assertions.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- This is a low-confidence assessment of the supplied static files; no code was run and capabilities, performance, or security were not independently verified.
- The README mentions local device processing and privacy but does not describe the complete data flow or protections for chat content, model requests, or credentials.
- The project is under active development; before deploying to real devices, consult the full build documentation and assess permissions and recovery for device-control actions.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.1.0 ESP-Claw v0.1.0
Release notes excerpt
Release Note - ESP-Claw v0.1.0 The first release of ESP-Claw. Thanks to all contributors for their code, board support, and feedback that helped shape this release. > [!TIP] > > For build and usage instructions, please refer to the documentation at https://esp-claw.com/ <details> <summary> <b>Architecture diagram</b> (Click here to expand) </summary> <img width="5196" height="6652" alt="architecture-diagram-dark-2" src="https://github.com/user-attachments/assets/5e989a59-0fea-4c29-8414-8c98e82ef4ab" /> </details> Agent Core …View on GitHub ↗