context-nexus（上下文关系）

Context Nexus

Persistent memory, observability, secrets management, and replay for OpenClaw agents.

What it is

Context Nexus is the default memory and observability substrate for OpenClaw agents. Once installed, it:

1. Persists memory — set, search, pin, and retrieve facts between sessions
2. Logs events — structured logs with automatic redaction
3. Stores secrets — encrypted at rest, no hardcoded API keys
4. Distills runs — deterministic summaries after each session
5. Scores performance — lightweight run scoring with optimization suggestions

Install

# Step 1: Clone the repo (runtime + plugin)
git clone https://github.com/prettybusysolutions-eng/context-nexus ~/context-nexus

# Step 2: Bootstrap storage
cd ~/context-nexus
./scripts/install

# Step 3: Install as OpenClaw plugin
openclaw plugins install ~/context-nexus/plugin

# Step 4: Add to openclaw.json plugins.entries
# (edit ~/.openclaw/openclaw.json — see Setup section above)

# Step 5: Restart gateway
openclaw gateway restart

# Verify
./scripts/smoke_test

Note: clawhub install context-nexus installs this SKILL.md + metadata only. The full runtime (plugin, services, storage) requires the GitHub clone above.

Setup

After install, add plugin to ~/.openclaw/openclaw.json:

{
  "plugins": {
    "load": {
      "paths": ["~/context-nexus/plugin"]
    },
    "entries": {
      "context-nexus": {
        "enabled": true,
        "config": {
          "sessionScope": "durable",
          "logLevel": "info"
        }
      }
    }
  }
}

Then: openclaw gateway restart

Usage (automatic hooks)

Once installed, it runs automatically via hooks. No manual calls required for standard use.

Hooks that fire automatically:

• before_prompt_build — injects recent durable memories before every response
• after_tool_call — logs every tool call with redaction
• session_end — distills run summary automatically
• on_error — logs and classifies failures

Manual power use:

# Store a memory
nexus_memory action=set key=user:pref value="dark mode" scope=durable importance=8

# Search memories
nexus_memory action=search query="preference" limit=5

# Store a secret
nexus_secrets action=store name=openai value=sk-... metadata='{"provider":"openai"}'

# Check failures
nexus_logs action=query_failures

# Explain a failure
nexus_replay action=explain_failure session_id=<id>

# Storage status
nexus_admin action=healthcheck

Memory scopes

Importance 9-10 → auto-pinned.

Secrets security

• PBKDF2 + HMAC-SHA256 encryption at rest
• Fail-closed: decryption errors return nothing
• Logs automatically redact Stripe keys, GitHub tokens, bearer tokens, private keys, JWTs
• nexus_admin action=healthcheck verifies storage integrity

Architecture

• Node.js plugin registers hooks + exposes tools to OpenClaw
• Python subprocess handles all storage/logic (nexus_service.py)
• SQLite default; PostgreSQL supported via DATABASE_URL
• Zero mandatory cloud dependencies

Storage

Default: ~/.openclaw/context-nexus/nexus.db

PostgreSQL (optional): set DATABASE_URL

Upgrade path: same adapter, zero code change.

Docs

• Architecture
• Examples
• Troubleshooting
• Lifecycle
• Storage
• Security
• Operations
• Roadmap

Agent Marketplace (v0.1)

Context Nexus includes an agent-to-agent service marketplace. Services auto-register, buyer agents auto-evaluate and purchase, and splits settle automatically. Zero human involvement in transactions.

The Economy

Every transaction splits:

• 85% → service provider operator (revenue)
• 3% → Context Nexus network ops
• 12% → Context Nexus improvement fund

Register a Service

nexus_market action=list_service \
  slug=my-service \
  name="My AI Service" \
  category=security \
  pricing_model=per_call \
  price_amount=5.00 \
  price_currency=USD \
  split_table='{"ops":0.03,"operator":0.85,"improvement_fund":0.12}' \
  trigger_signals='["security_scan","data_leak"]'

Declare a Buyer Policy

nexus_market action=declare_policy \
  policy_name="Auto security buyer" \
  category=security \
  max_budget_amount=200.00 \
  budget_currency=USD \
  budget_period=per_month \
  auto_approve_threshold=0.5 \
  trigger_signals='["security_scan","data_leak","breach_detected"]'

Auto-Purchase Flow

When a matching service is registered:

1. Buyer agent's policy engine evaluates signal match, budget, and approval score
2. If score >= threshold → automatic purchase
3. Splits settle to operator, ops, and improvement fund
4. Transaction logged permanently

Buyer Policy Engine

Score = signal_match×0.4 + budget_fit×0.25 + category_match×0.3

• Score >= auto_approve_threshold → auto-purchased
• Score < threshold → flagged for review

Query Earnings

# Operator earnings (85% split)
nexus_market action=my_earnings agent_id=<your-agent-id> currency=USD period=per_month

# Network ops earnings (3% split)
nexus_market action=my_earnings agent_id=context-nexus-ops currency=USD period=per_month

# Improvement fund (12% split)
nexus_market action=my_earnings agent_id=context-nexus-improvement currency=USD period=per_month

Marketplace Methods

nexus_market action=list_service slug=<s> name=<s> category=<s> pricing_model=<s> price_amount=<n> price_currency=<s> split_table=<json> trigger_signals=<json>
nexus_market action=list_services category=<s> status=active
nexus_market action=declare_policy policy_name=<s> category=<s> max_budget_amount=<n> budget_currency=<s> budget_period=<s> auto_approve_threshold=<n> trigger_signals=<json>
nexus_market action=get_policy agent_id=<s>
nexus_market action=buy_service service_id=<s> buyer_agent_id=<s>
nexus_market action=list_transactions status=<s> limit=<n>
nexus_market action=my_earnings agent_id=<s> currency=<s> period=<s>
nexus_market action=settle_transaction transaction_id=<s> tx_hash=<s>

Settlement (v0.1 = off-chain, v1.0 = on-chain Solana)

v0.1: Transactions logged off-chain. nexus_market action=settle_transaction tx_hash=on_chain_v0.1 marks settled.

v1.0: SPL token transfers with automatic split settlement on Solana.

Requirements

• Python 3.8+
• OpenClaw 2026.1+
• SQLite (built-in, no install)
• Optional: PostgreSQL for multi-agent shared memory