BeaconTranslation site

What is Beacon?

Beacon is an AI programming agent memory and execution logging tool developed by Asymptote Labs. Its official website is beacon.sh, and its open-source code is available at Asymptote-Labs/agent-beacon under the MIT License. It connects to more than 20 runtime environments, including Claude Code, Cursor, Codex, and OpenCode, creating traceable records of scattered sessions and passing on useful insights to subsequent agents.Official Documentation.;Project Repository

For developers who frequently switch between programming tools, the practical value lies in reducing the need to repeatedly explain the project context, as well as making it easier to review the commands, troubleshooting steps, and attempts involved in a particular modification. It is suitable for individual developers maintaining long-term codebases, engineering teams that need to review agent behavior, and organizations seeking to centrally manage multiple programming assistants.

Key Features and Use Cases

• Unified Session Log: Organize the activity data exposed by different assistants into a common record structure for viewing history, troubleshooting workflows, and analyzing usage. Since the data fields provided by each tool differ, “supported integration” should not be interpreted to mean that their data collection capabilities are completely identical.
• Formed as a memory after review: Candidate experiences are proposed from selected sessions; only after reviewing the original records, supplementing the experience text, and confirming the details are they added to the project memory. This approach is suitable for preserving verified debugging methods and project conventions, and helps avoid treating failed attempts as correct answers.Memory Mechanisms
• Reusing Skills and Handing Off Tasks: Approved memories can be queried by the MCP or converted into Agent Skills. Session handoffs can also generate task summaries, allowing another assistant to continue handling the paused work—useful when quotas are exhausted, when switching tools, or when rechecking stuck tasks.Skill Description.;Handoff of the Conversation

How to Get Started

1. Select the macOS, Windows, or Linux installation method from the official installation guide, and check the Beacon version after installation.
2. 用 beacon endpoint install --harness claude,codex,cursor Configure the assistant you'll actually be using, and carefully choose between "Local" and "Beacon Cloud." If you only want to save data locally, be sure to select "Local."
3. Complete a small task in the configured assistant, then run Beacon Endpoint Status To check the collection status, use beacon endpoint dashboard --open View the conversation.
4. Start by documenting lessons learned from a successful mission; after review, apply them through MCP or by reusing skills. First, verify that your recollection is accurate, then expand it to the team’s codebase.Getting Started Quickly

Price and Usage Limits

Open-source code and hosting services should be considered separately. The Beacon Cloud Free plan offers session retrieval and 50 “findings”—that is, insights regarding risks, skills, or usage patterns; the Pro plan, which costs $20 per month when paid annually, provides an unlimited number of findings, cloud storage, and more comprehensive MCP features. The Team plan is charged per seat, but there are discrepancies between the pricing card and the comparison table on the official website; be sure to confirm the final invoice before purchasing. For the Enterprise plan, please request a quote.Official Price

Local mode does not automatically forward data, while cloud mode sends the selected telemetry information. Interactive installation preselects the cloud mode; you should verify the target account and data collection scope before confirming. Calling external evaluation models also generates separate data transfers; if your organization prohibits outbound data transmission, you should skip external evaluation or configure an appropriate internal service. Storage is isolated by project by default; local and cloud storage do not automatically synchronize with each other.Deployment Steps.;Privacy and Project Scope

How to Choose Among Similar Tools

The decision should be based on whether a complete work history and manual review are necessary, rather than on the assumption that the more entries, the better. Stable project specifications should still be stored in version control, while records related to temporary issues and outdated solutions should be periodically purged.

Recent Updates

Version 1.3.52, released on November 1, 2026, fixed recurring issues with the redaction of sensitive information and added or improved features such as secondary telemetry backend forwarding and permission protection for Windows credential files.v1.3.52 Release Notes