Memory, Context & Sessions

Claw Is Not Stateless

Claw does not behave like a fresh bot on every message.

It keeps continuity through a mix of:

• Telegram conversation history
• app-level conversation state
• a warm per-user runtime when available
• a Claude session that can be resumed across messages

For power users, that means you can do iterative work instead of rebuilding context every time.

What This Feels Like In Practice

• follow-up questions can stay compact
• file-based workflows can continue across messages
• Claw can keep working in the same operating context instead of starting from zero
• ongoing reviews of agents, balances, and positions feel more like a session than a support bot

Important Reality

Claw uses persistent runtime behavior when possible. It is not best described as an "ephemeral sandbox per message."

That matters because:

• context can carry forward
• generated files may still matter in later turns
• the runtime may be resumed instead of recreated

When State Changes

Session continuity is strong, but not absolute.

State can change when:

• the runtime is recreated
• the underlying template changes
• account auth changes require runtime migration
• an earlier session is no longer available

The right user expectation is: Claw usually feels persistent, but it is still a managed runtime rather than your own always-on shell.

How To Use That

For power users, this changes how you prompt:

• you can build on prior work
• you can ask for revisions to generated outputs
• you can keep a thread focused on one operational task

If you want the user-level knobs on top of this runtime, see Models & Preferences.