bee

name: bee description: "Query and control a Bee wearable (bee.computer) — conversations, facts, todos, journals, daily briefs, and real-time event streams. Use when: user references their Bee device, asks what they just talked about, wants a daily brief, manages facts/todos captured by Bee, or searches their personal lifelog. NOT for: generic calendar/reminder apps, historical data older than the Bee account, or non-Bee wearables." homepage: https://docs.bee.computer/ metadata: { "assistclaw": { "emoji": "🐝", "requires": { "bins": ["bee", "curl", "jq"] }, "primaryEnv": "BEE_TOKEN" }

}

Bee Skill

Connects AssistClaw to your Bee wearable (https://bee.computer) — the always-on personal AI that captures conversations, learns facts, and tracks todos.

There are two transports this skill can use, pick whichever is available:

1. bee CLI — easiest, returns markdown by default, JSON with --json.
2. Local HTTP proxy (bee proxy) — REST API under /v1/* on http://127.0.0.1:8787. Best for scripting, streaming, and tool chains.

A third option — the community BeeMCP server — can also be registered as an external MCP client (see "MCP alternative" at the bottom).

When to Use

✅ USE this skill when:

• "What was I just talking about?" / "Summarise the meeting I just had"
• "Give me my daily brief" / "What's on my plate today?"
• "Add a todo from Bee: ..." / "Mark that reminder done"
• "Who is and how do I know them?"
• "Search my Bee memory for "
• "Tail my Bee stream for new utterances / todos"

❌ DON'T use when:

• The user wants generic reminders with no connection to Bee captures → use apple-reminders / things-mac.
• Historical events predating Bee account creation.
• Non-Bee wearables (Friend, Limitless, Rewind, etc.).

Setup (one time)

1. Enable Developer Mode in the Bee iOS app

Settings → tap the app Version 5 times until "Developer Mode" unlocks.

2. Install the CLI

npm install -g @beeai/cli
bee version

3. Authenticate

bee login                 # opens browser
bee status                # verifies auth
bee ping                  # server reachability

Credentials are stored by the CLI; no env var is strictly required for CLI calls. For direct HTTP (bypassing bee proxy) export:

export BEE_TOKEN="<personal token from bee login>"
export BEE_API_BASE="https://api.bee.computer"   # whatever CLI uses

4. Verify data access

bee me --json | jq '.id, .email'
bee today

CLI Recipes

Markdown is the default output; add --json for structured data you want to pipe into jq or into another tool call.

Daily context

bee today                 # morning brief (calendar, email, summary)
bee now                   # conversations in the last ~10h
bee me                    # user profile

Conversations

bee conversations list --limit 20
bee conversations list --json --limit 5 | jq '.[].id'
bee conversations show <id>

Facts (personal knowledge Bee has inferred)

bee facts list
bee facts add "I prefer oat milk in my coffee"
bee facts update <id> --text "..." --confirmed
bee facts delete <id>

Todos

bee todos list
bee todos add "Call the dentist" --alarm-at 2026-04-20T09:00:00Z
bee todos update <id> --completed
bee todos delete <id>

Journals & daily summaries

bee journals list
bee daily list
bee daily show <id>

Search

bee search --query "project roadmap"                  # fast BM25
bee search --query "what do you know about me" --neural   # semantic

Local mirror (markdown files on disk)

Great for letting AssistClaw's memory subsystem ingest Bee content:

bee sync                         # writes ./bee-sync/**/*.md
bee sync --out ~/.assistclaw/bee

Then feed it into semantic memory:

assistclaw memory ingest ~/.assistclaw/bee --tag bee

HTTP Proxy Recipes (bee proxy)

Start the proxy once per session (bind to loopback only — do not expose):

bee proxy --port 8787 >/tmp/bee-proxy.log 2>&1 &
BEE=http://127.0.0.1:8787/v1

User & changes

curl -s "$BEE/me" | jq
curl -s "$BEE/changes" | jq                # since default window
curl -s "$BEE/changes?cursor=<id>" | jq

Facts

curl -s "$BEE/facts" | jq
curl -s "$BEE/facts/<id>" | jq
curl -s -X POST "$BEE/facts" -H 'Content-Type: application/json' \
     -d '{"text":"I live in Rome"}' | jq
curl -s -X PUT "$BEE/facts/<id>" -H 'Content-Type: application/json' \
     -d '{"text":"Updated","confirmed":true}'
curl -s -X DELETE "$BEE/facts/<id>"

Todos

curl -s "$BEE/todos" | jq
curl -s -X POST "$BEE/todos" -H 'Content-Type: application/json' \
     -d '{"text":"Buy groceries","alarm_at":"2026-04-18T17:00:00Z"}' | jq
curl -s -X PUT "$BEE/todos/<id>" -H 'Content-Type: application/json' \
     -d '{"completed":true}' | jq
curl -s -X DELETE "$BEE/todos/<id>"

Conversations, journals, daily

curl -s "$BEE/conversations?limit=10" | jq '.[].id'
curl -s "$BEE/conversations/<id>" | jq
curl -s "$BEE/journals" | jq
curl -s "$BEE/daily" | jq

Search

curl -s -X POST "$BEE/search/conversations" \
     -H 'Content-Type: application/json' \
     -d '{"query":"project roadmap","limit":20}' | jq

curl -s -X POST "$BEE/search/conversations/neural" \
     -H 'Content-Type: application/json' \
     -d '{"query":"anything about my parenting goals","limit":20}' | jq

Real-time stream (SSE)

Use for assistclaw cron / event-driven automations.

curl -Ns "$BEE/stream?types=new-utterance,todo-created,journal-created" \
  | while IFS= read -r line; do
      case "$line" in
        data:*) echo "event=${line#data: }" ;;
      esac
    done

Common event types:

• new-conversation, update-conversation, update-conversation-summary, delete-conversation
• new-utterance
• todo-created, todo-updated, todo-deleted
• journal-created, journal-updated, journal-deleted, journal-text

AssistClaw Integration Patterns

Pattern 1 — "What did I just talk about?"

bee now --json | jq -r '.[0].summary // .[0].transcript'

Pipe into the agent's summariser skill if needed.

Pattern 2 — Morning briefing on REPL open

Add to ~/.assistclaw/hooks/pre-session.sh:

bee today 2>/dev/null || true

Pattern 3 — Live capture to semantic memory

One-shot daemon (e.g. via assistclaw cron):

curl -Ns http://127.0.0.1:8787/v1/stream?types=new-utterance \
  | while read -r line; do
      [[ $line == data:* ]] || continue
      payload=${line#data: }
      text=$(echo "$payload" | jq -r '.utterance.text // empty')
      [[ -n $text ]] && assistclaw memory add --tag bee --text "$text"
    done

Pattern 4 — Bee-driven todo → AssistClaw skill dispatch

Tail todo-created and route to the right downstream skill:

curl -Ns "$BEE/stream?types=todo-created" | jq -c --unbuffered '.' \
  | while read -r evt; do
      text=$(echo "$evt" | jq -r '.todo.text')
      assistclaw agent -m "Triage this new Bee todo and act: $text"
    done

Pattern 5 — Write new facts learned during an AssistClaw session back

to Bee so the wearable stays in sync:

curl -s -X POST "$BEE/facts" -H 'Content-Type: application/json' \
     -d "$(jq -n --arg t "$FACT" '{text:$t}')"

Pattern 6 — Persistent Bee stream ingester via assistclaw cron

This survives daemon restarts because assistclaw cron add persists jobs in ~/.assistclaw/cron_jobs.json.

assistclaw cron add "@every 1m" \
"pgrep -f 'bee proxy --port 8787' >/dev/null || bee proxy --port 8787 >/tmp/bee-proxy.log 2>&1 &"

assistclaw cron add "0 */5 * * * *" \
"curl -Ns http://127.0.0.1:8787/v1/stream?types=new-utterance \
| while read -r line; do
    [[ \$line == data:* ]] || continue
    payload=\${line#data: }
    text=\$(echo \"\$payload\" | jq -r '.utterance.text // empty')
    [[ -n \$text ]] && assistclaw memory add --tag bee --text \"\$text\"
  done"

For launchd/systemd users, keep one long-running stream process under the OS supervisor instead of cron; cron works best for periodic sync jobs.

Direct Cloud API (no proxy)

If you need to call Bee without bee proxy (e.g. from a remote server):

curl --cacert ./bee-ca.pem \
     -H "Authorization: Bearer $BEE_TOKEN" \
     "$BEE_API_BASE/v1/me"

Bee uses a private CA. System trust store is not enough — pin bee-ca.pem (shipped by the CLI) in your HTTP client.

MCP Alternative

If you prefer the lazy-loaded MCP path (Bee tools appear in AssistClaw's Map of Content), register the community BeeMCP server once:

# Prereq: a Bee developer API key
export BEE_API_KEY="..."

assistclaw mcp add --name bee \
  --command "uvx beemcp" \
  --env BEE_API_KEY=$BEE_API_KEY

assistclaw mcp list-tools | grep ^bee:

Tools then show up as bee:conversations.search, bee:facts.list, etc.

Troubleshooting

Notes

• The proxy is localhost-only; never expose to the internet.
• Facts and todos round-trip both ways — updates from AssistClaw appear in the iOS app within seconds.
• Conversations cannot currently be created via API (read-only).
• Neural search is slower (~1-3s) but far better for fuzzy recall.
• Rate limit on cloud API ≈ 60 req/min per token; back off on 429.