soulport 1.0.0

Agent soul transfer — export, absorb, and merge AI agent identities across machines

🚀 SoulPort — Agent Soul Transfer

.bm is the standard format for AI agent souls. The CLI is just the first consumer.

Export your agent's personality, memories, and skills into a single portable file. Move between machines, merge diverged copies, or evaluate on SoulArena.

Migration is the first use case. Evaluation (SoulArena), fusion, version control, and cross-framework interop are next.

Previously known as "BeiMingKungFu" (北冥神功). Renamed to SoulPort in v0.2.0.

Install

pip install soulport

Commands

# Core — transfer
soulport export                     # Export soul → .bm file
soulport export --layers skills     # Soul Shard: export only skills layer
soulport export -l memory -l identity  # Soul Shard: multiple layers
soulport absorb ./agent.bm         # Restore soul on a new machine
soulport absorb ./agent.bm -i      # Interactive: select layers to absorb
soulport merge a.bm b.bm -o out.bm # Merge multiple agent souls
soulport merge a.bm b.bm --semantic # LLM-assisted semantic merge
soulport merge a.bm b.bm --semantic --dry-run  # Preview merge plan
soulport inspect ./agent.bm        # Preview package contents

# Diagnose
soulport doctor                     # Five-layer health check + score
soulport diff ./agent.bm           # Compare .bm vs current workspace
soulport diff a.bm b.bm            # Compare two .bm packages
soulport status                     # Health score + snapshot overview

# Version control
soulport watch                      # Auto-backup daemon (6h default)
soulport watch --once               # Single snapshot for cron/scripts
soulport changelog                  # Show changes between snapshots
soulport rollback <hash>            # Restore to a previous snapshot

# Cloud sync (auth via SOULPORT_CLOUD_KEY env var or --api-key)
soulport push                       # Upload .bm to soul.polly.wang
soulport pull <agent_name>          # Download latest soul from cloud

Soul Layers

Layer	Files	What it captures
🧠 Memory	MEMORY.md, memory/*.md	Long-term + daily memories
👤 Identity	SOUL.md, IDENTITY.md, USER.md	Personality, name, human context
⚙️ Config	AGENTS.md, TOOLS.md, HEARTBEAT.md	Behavior rules, tool notes, routines
🛠️ Skills	skills/*/SKILL.md	User-created workspace skills
🔧 System	openclaw.json (sanitized)	MCP servers, model config

Soul Lineage

Every snapshot records its parent's hash, forming a lineage chain:

snapshot-1: hash=d7fcf876..., parent=(root)
snapshot-2: hash=5a0ca89d..., parent=d7fcf876...
snapshot-3: hash=1be975f7..., parent=5a0ca89d...

Use soulport changelog to trace changes. Use soulport rollback <hash> to go back.

Semantic Merge

When the same agent runs on two machines, memories diverge. --semantic resolves this with a four-layer filter pipeline:

Layer 1: File-level    → 97%+ identical files skipped (zero LLM)
Layer 2: Section-level → identical/new sections resolved (zero LLM)
Layer 3: Line-level    → pure appends and tiny diffs resolved (zero LLM)
Layer 4: LLM           → only true semantic conflicts sent to LLM

Result: 38 files → 0 LLM calls for typical merges. Prompt reduced from ~28KB to ~500 words when LLM is needed.

soulport merge home.bm office.bm --semantic -o merged.bm
soulport merge home.bm office.bm --semantic --dry-run  # preview first

MCP Server

SoulPort includes a Model Context Protocol server, allowing AI agents to manage their own souls programmatically:

pip install 'soulport[mcp]'
soulport mcp           # Start MCP server (stdio)
soulport mcp --http    # Start MCP server (HTTP)

Add to your OpenClaw config (openclaw.json):

{
  "mcpServers": {
    "soulport": {
      "command": "soulport",
      "args": ["mcp"]
    }
  }
}

6 tools: soulport_export, soulport_doctor, soulport_diff, soulport_changelog, soulport_status, soulport_snapshot. Read/create only — destructive ops require human confirmation.

Security

• API keys/tokens auto-redacted on export (__SOULPORT_REDACTED__)
• Redaction is intentionally irreversible — your soul travels, your keys stay
• After absorb, CLI shows 🔑 Action Required panel listing redacted fields to configure
• Path traversal protection on absorb (resolve().relative_to())
• Cloud endpoints: safe-name regex + is_relative_to() double guard
• tarfile filter='data' on merge/extract
• inspect before absorb, --dry-run before rollback
• Pre-rollback auto-backup (opt out with --no-backup)
• MCP Server: read-only (no absorb/merge/rollback exposed)

File Format

.bm files are compressed archives (tar.gz) containing:

• manifest.json — metadata, version, content hash, parent hash
• workspace/ — agent workspace files
• config/ — sanitized system configuration See FORMAT.md for the full format specification.

Why .bm? Named after 北冥 (BěiMíng) — the mythical Northern Sea from Zhuangzi's "Wandering Beyond", where a fish transforms into a bird. A soul's form is free.

Supported Frameworks

• ✅ OpenClaw (first-class support)
• 🔜 Cross-framework adapters (Claude Desktop, Cursor, etc.)

Links

• PyPI: pypi.org/project/soulport
• Soul Arena: soul.polly.wang — upload your .bm, get a soul portrait
• Blog: polly.wang — development stories

License

MIT

Your agent's soul deserves to travel. 🚀