bamdra-user-bind

The identity and living profile layer for the Bamdra suite.

It can run independently, and it is also auto-provisioned by bamdra-openclaw-memory.

Install directly:

openclaw plugins install @bamdra/bamdra-user-bind

Release package:

• GitHub Releases: https://github.com/bamdra/bamdra-user-bind/releases
• You can also build a local release bundle with pnpm package:release

中文文档

What it does

bamdra-user-bind turns raw channel sender IDs into a stable user boundary.

It also becomes the user's evolving profile layer, including:

• userId-scoped preferred address
• timezone
• tone preferences
• role
• long-lived user notes

Recent behavior improvements:

• channel-scoped user IDs now make profile ownership explicit across Feishu, Telegram, WhatsApp, Discord, Google Chat, Slack, Mattermost, Signal, iMessage, and Microsoft Teams
• when a stable binding is temporarily unavailable, the runtime can persist a provisional profile first and merge it into the stable profile later
• profile updates now support semantic replace, append, and remove instead of only blind overwrite
• the Markdown mirror now keeps frontmatter as the machine-readable source and renders a synchronized Confirmed Profile section for humans

Profile Policy

• userId is the primary key for personalization
• preferred address should live in the bound profile, not in scattered workspace USER.md files
• workspace USER.md should stay minimal and only keep environment facts
• if the current turn explicitly asks for a different address, follow the current turn
• admin tools are for repair, merge, audit, and sync, not for blind bulk rewriting

Why it matters

Without an identity layer:

• the same person can fragment across channels or sessions
• memory can attach to the wrong boundary
• personalization becomes fragile

With it:

• user-aware memory becomes stable
• personalization survives new sessions
• the assistant can gradually adapt to the user's style and working habits

Storage model

• primary store: ~/.openclaw/data/bamdra-user-bind/profiles.sqlite
• editable Markdown mirrors: ~/.openclaw/data/bamdra-user-bind/profiles/private/{userId}.md
• export directory: ~/.openclaw/data/bamdra-user-bind/exports/

The SQLite store is the controlled source of truth.

The Markdown mirror is for humans, so profiles stay editable like a living per-user guide instead of becoming a hidden black box.

Profile update semantics

Not every durable user preference should overwrite the entire old field.

bamdra-user-bind now distinguishes between:

• replace: the user is correcting or changing an earlier durable preference
• append: the user is adding another durable preference without revoking the old one
• remove: the user explicitly wants one older durable trait removed

This is especially useful for fields like preferences, personality, and notes, where the right behavior is often incremental.

Provisional identity and later merge

Some channels or app states can temporarily fail to resolve a stable bound identity on the first turn.

In those cases, the runtime can:

• persist a provisional profile immediately so the user's newly stated preferences are not lost
• keep trying to repair the binding in the background
• merge the provisional profile into the stable user profile once the real binding is available

This keeps the user experience responsive without turning missing identity resolution into data loss.

Best practice

• keep SQLite local
• keep profile mirrors private
• let humans edit the mirror gradually
• use admin tools only for audit, merge, repair, and maintenance
• when updating how someone is addressed, update the bound profile for that userId first

Architecture

What it unlocks

With bamdra-openclaw-memory:

• memory becomes user-aware instead of session-only

With bamdra-memory-vector:

• private notes stay private while still influencing local recall

Repository

• GitHub organization
• Repository
• Releases