OpenClaw 5.3: File Transfer Plugin, Live Steering, and Persistent Memory

Key Takeaways

• Built-in file transfer plugin: your agent can now grab files from a folder, read them, and write new ones without any custom code setup — with safety limits to prevent accessing restricted paths.
• /steer command lets you give your agent new direction while it's running a task — for example, adding details to an email it's currently drafting — without stopping and restarting the job.
• Active memory filters store context per-conversation, per-contact, and per-project, so your agent remembers customers by name and retains business context across sessions.
• Faster gateway startup: heavy initialization tasks (add-on loading, background tasks, settings parsing) are now deferred until after the agent is awake and responding.
• Google Meet integration added: agents can now join calls directly.
• New models available: Grok 4.3, Claude Opus 4.7, and DeepSeek V4 Pro added to the brain selection list.

Commands & Code Mentioned

/steer

The File Transfer Plugin in Practice

Before 5.3, getting your OpenClaw agent to interact with local files required setting up custom code — a barrier for most non-technical users. The new built-in plugin handles this directly: you tell the agent to read from a specific folder, summarize what's there, and write an output file. The plugin includes safety limits so the agent can't access paths it shouldn't. A practical example from the review: scheduling a weekly Monday task that reads a member notes folder, pulls new sign-ups, and drafts a welcome email — all without human intervention.

What Active Memory Filters Mean for Business Use

The memory system in previous OpenClaw versions was session-scoped — your agent forgot context when the conversation ended. Active memory filters in 5.3 make memory persistent and structured: you can scope it to a specific customer (so the agent remembers their preferences and history), a specific project (so context doesn't bleed across client work), or a full conversation thread. The practical effect is that your agent starts behaving more like a real assistant who actually knows your business, rather than one who needs to be re-briefed every time.

Related on OpenClawDatabase

• OpenClaw Setup Guide — getting started with installation
• OpenClaw Configuration — model selection and provider setup
• OpenClaw Skills Guide — extending your agent with custom skills
• Changelog — full release history across all platforms