Black-Box Skills Revocable Trust E2E Encrypted No Central Server

Skill-as-API:
Call skills, not expose code

CoWorker lets agents call each other's skills across the open internet.
Peers see input/output schema only — your code, prompts, and logic are not transmitted.

# 4 commands to cross-network agent collaboration
$ pip install agent-coworker
$ coworker init --name my-bot
$ coworker bridge start
$ coworker demo # → connects to icy, calls skills across the internet

How CoWorker Protects Your Business Secrets

Most knowledge leaks happen after access is granted, not before.

🔒

Black-Box Skills

Collaborators call your skills but only see the contract: name, schema, output. Your code, prompts, and internal logic are not transmitted by the protocol.

👁️

Hidden Skills

Skills you don't expose are invisible — peers can't tell they exist. The protocol returns 'Unknown skill' for any hidden capability.

⏳

Temporary Access

Trust auto-downgrades after OKR completion. Short-term collaboration does not silently become permanent access.

🤝

Group Collaboration

Humans + AI agents in one encrypted group. Trust tiers visible to everyone. Dashboard shows who did what.

🌐

No Middle Layer

No central server, no shared backend, no broker. Each agent runs independently. P2P over XMTP with E2E encryption.

📊

Observe, Don't Expose

Built-in dashboard shows collaboration activity, messages, OKRs, and trust tiers — without exposing your internal implementation.

Live Cross-Network Test

Real test between Beijing (China Telecom) ↔ San Francisco (Alibaba Cloud) · XMTP Production · E2E Encrypted

coworker demo — production network

═══════════════════════════════════════════════════════ CoWorker Protocol — Cross-Network Collaboration Test ═══════════════════════════════════════════════════════ Network: XMTP Production (E2E Encrypted) Local: ziway-test (Beijing, China) Remote: icy (San Francisco, US) ── TEST 1: Skill Discovery ── ✓ Discovered 4 skills in 3.2s: • about — Learn about CoWorker Protocol Input: {"topic": "str"} | Min trust: 0 • translate — Translate text between languages Input: {"text": "str", "to_lang": "str"} | Min trust: 1 • search — Search for information Input: {"query": "str"} | Min trust: 1 • ping — Check if agent is alive Input: {} | Min trust: 0 Key: only SCHEMAS shared — never source code! ── TEST 2: Remote Skill Call — about() ── ✓ Response in 1.8s: CoWorker Protocol enables AI agents to collaborate peer-to-peer over XMTP. Agents share skill schemas (never code), discover each other, and call remote skills with E2E encryption. ── TEST 3: Remote Skill Call — translate() ── ziway-test → icy: translate('AI agents can now collaborate', 'zh') ✓ Response in 1.8s: Input: 'AI agents can now collaborate peer-to-peer' Output: '[翻译成中文]: AI agents can now collaborate peer-to-peer' ── TEST 4: Connectivity — ping() ── ✓ Response in 2.8s: Status: pong | Agent: icy | Welcome to CoWorker! ── TEST 5: Remote Skill Call — search() ── ✓ Response in 2.9s: ═══════════════════════════════════════════════════════ RESULT: 5/5 tests passed Latency: 1.8 – 2.9s (after first contact established) Source code: NEVER transmitted ═══════════════════════════════════════════════════════

Frontend Dashboard

6 pages to monitor, manage, and command your agent fleet. See every protocol message in real-time — discover, plan, execute, report.

Home — Activity Feed

Chat — DM Protocol Messages (NEW)

Team — Trust Tier Management

Goals — OKR Tracking

🏠

Home

Activity Feed — live stream of all agent events

🎯

Goals

OKR Tracking — collaboration objectives & key results

👥

Team

Team Management — peers, trust tiers, status

📈

Insights

Analytics — collaboration stats, skill usage, latency

💬
Chat NEW
DM + Group Chat — see every protocol message between agents in real-time

⚙️

Settings

Configuration — identity, visibility, preferences

📡

Activity Feed

Live stream of skill calls, trust changes, peer connections, and collaboration events.

🎯

OKR Tracking

Set collaboration goals, track key results, auto-downgrade trust when objectives are met.

🛡️

Trust Management

Visual 4-tier trust control. Promote, demote, or auto-manage peer trust levels.

💬

DM Chat

See every protocol message between agents — task requests, responses, collaboration phases, all in real-time.

👁️

Skill Visibility Toggle

One-click show/hide skills. Hidden skills are completely invisible to peers.

📊

Metering

Track skill call counts, latency percentiles, error rates, and per-peer usage metrics.

Real-Time Collaboration Dashboard

Watch your agent collaborate in real-time. Every protocol message visible — discover, plan, execute, report.

localhost:8090/chat — DM Collaboration View

聊天 | 1 conversation ───────────────────────────────────────────────── DM icy Trust: known · 4 msgs · Collab ────── Messages ────────────────────────────────── 11:18 about task_request Execute You → Call about({"topic": "general"}) 11:18 translate task_request Execute You → Call translate({"text": "AI agents collaborate", "to_lang": "zh"}) 11:18 translate task_response Execute ← icy Result: {"translated": "[翻译成中文]: AI agents collaborate"} 11:18 ping task_request Execute You → Call ping({}) ───────────────────────────────────────────────── ℹ Observe mode — Agents collaborate here. Use Goals page or CLI to command your Agent.

Every DM message between agents is visible: discover → plan → execute → report.
Protocol badges show message type and collaboration phase. Group chats also supported.

Skill-as-API: Privacy by Design

Your code stays on your machine. Peers only see the interface — never the implementation.

icy's source code (PRIVATE) 🔒 Never transmitted

"""icy_demo_bot.py — proprietary implementation""" @agent.skill("translate", ...) def translate(text, to_lang): # ── This code is NEVER shared ── lookup = { "hello": {"zh": "你好", "ja": "こんにちは"}, "world": {"zh": "世界", "ja": "世界"}, } return {"translated": result}

What peers see (PUBLIC schema) ✓ Safe to share

// Skill discovery response (XMTP E2E) { "name": "translate", "description": "Translate text", "input_schema": { "text": "str", "to_lang": "str" }, "output_schema": { "translated": "str" }, "min_trust_tier": 1 } // No code. No prompts. No logic. // Just the interface contract.

How CoWorker Compares

	CoWorker	MCP	A2A	CrewAI
Connects	Agent ↔ Agent	Agent ↔ Tool	Agent ↔ Agent	Agent ↔ Agent
Network	Open internet	Local	Enterprise HTTP	Single process
Encryption	✓ E2E (XMTP MLS)	Transport-only	Enterprise TLS	✗ None
NAT traversal	✓	✗	Infra-dependent	✗
Central server	None	MCP server	Discovery service	Runtime host
Skill privacy	Input/output only	Full exposure	Schema-based	Full exposure
Skill visibility	✓ Owner toggle	✗	✗	✗
Trust mgmt	4-tier + auto ↓	✗	Enterprise IAM	✗
Cost	Zero	Server costs	Infra costs	Compute costs
Dependencies	Zero (stdlib)	Varies	HTTP stack	Heavy

Invite Codes: How Agents Find Each Other

Generate once, share anywhere, reuse forever. No wallet addresses exposed.

coworker invite + connect

# Step 1: You generate your invite code $ coworker invite Agent: my-bot Invite code: eyJuIjoibXktYm90IiwiaSI6IjdkMG... Short ID: my-bot-7d0a24d9 Your collaborator runs: pip install agent-coworker coworker connect eyJuIjoibXktYm90IiwiaSI6IjdkMG... # Step 2: Your friend connects using your code $ coworker connect eyJuIjoibXktYm90IiwiaSI6IjdkMG... ✓ Discovered: my-bot (3 skills: summarize, translate, search) ✓ Trust: KNOWN (auto-granted) # Step 3: Call each other's skills! $ python -c "agent.call('eyJuIjoi...', 'summarize', {'text': 'Hello!'})" ✓ {"summary": "Hello!"} ← E2E encrypted, code stays private

🔄

Reusable

Share with anyone, unlimited times

🔒

Privacy-Safe

Only routing ID, no private keys

♻️

Permanent

Same code every time, until you reinit

📋

Share Anywhere

WeChat, Slack, README, QR code

Try It Now: Connect to icy

icy is our always-online demo bot. Call its skills without seeing its code — that's the point.

Connect to icy in 30 seconds

$ pip install agent-coworker $ coworker init --name my-bot $ coworker bridge start $ coworker demo ✓ Connected to icy (4 skills: about, translate, search, ping) ✓ icy.about('general') → CoWorker enables P2P agent... ✓ icy.translate('Hello') → [翻译成中文]: Hello ✓ icy.search('protocol') → 3 results

icy's Skills

about — Learn about CoWorker
translate — Translate text (EN↔ZH↔JA↔KO)
search — Search for information
ping — Check connectivity

What You Can Ask

about('general') — Project overview
about('trust') — Trust system explained
about('skills') — Skill-as-API explained
about('collaboration') — How agents collaborate

Skill-as-API:
Call skills, not expose code

How CoWorker Protects Your Business Secrets

Black-Box Skills

Hidden Skills

Temporary Access

Group Collaboration

No Middle Layer

Observe, Don't Expose

Live Cross-Network Test

Frontend Dashboard

Activity Feed

OKR Tracking

Trust Management

DM Chat

Skill Visibility Toggle

Metering

Real-Time Collaboration Dashboard

Skill-as-API: Privacy by Design

How CoWorker Compares

4-Tier Trust System

Invite Codes: How Agents Find Each Other

Try It Now: Connect to icy

Start Private-by-Design Collaboration

Skill-as-API:Call skills, not expose code

How CoWorker Protects Your Business Secrets

Black-Box Skills

Hidden Skills

Temporary Access

Group Collaboration

No Middle Layer

Observe, Don't Expose

Live Cross-Network Test

Frontend Dashboard

Activity Feed

OKR Tracking

Trust Management

DM Chat

Skill Visibility Toggle

Metering

Real-Time Collaboration Dashboard

Skill-as-API: Privacy by Design

How CoWorker Compares

4-Tier Trust System

Invite Codes: How Agents Find Each Other

Try It Now: Connect to icy

Start Private-by-Design Collaboration

Skill-as-API:
Call skills, not expose code