mdx2025/BrainX V6

BrainX V6

Use BrainX when an OpenClaw agent needs durable memory, semantic recall, cross-session context, shared knowledge, or memory diagnostics.

Runtime Split

• BrainX skill / CLI: durable memory, retrieval, lifecycle, doctor, cron jobs, knowledge sync, backup/restore, evaluations.
• BrainX plugin: optional OpenClaw runtime bridge for wiki digest, working memory, JIT recall, tool advisories, failure capture, and legacy bridge fallback.
• LLM: reasoning layer. BrainX supplies context; it does not replace judgment.

Trust Model

BrainX memory is advisory.

• Prefer live evidence from code, logs, tests, runtime, database state, screenshots, and direct user corrections.
• Use verified memories first.
• Treat hypotheses and changelog memories as context, not authority.
• Exclude obsolete memories from bootstrap and recall surfaces.
• Keep write-path runtime features disabled until explicitly enabled.

Setup

Required environment:

DATABASE_URL=postgresql://brainx:[email protected]:5432/brainx
OPENAI_API_KEY=your-openai-api-key
OPENAI_EMBEDDING_MODEL=text-embedding-3-small
OPENAI_EMBEDDING_DIMENSIONS=1536

Install:

npm install
chmod +x brainx
psql "$DATABASE_URL" -f sql/v3-schema.sql
for file in sql/migrations/*.sql; do psql "$DATABASE_URL" -f "$file"; done
./brainx doctor --full

Core Commands

./brainx add "memory text" --type note --context project:example --importance 6
./brainx search --query "what should I remember?" --limit 5
./brainx inject "task context to enrich"
./brainx doctor --full --json
./brainx fix
./brainx metrics
./brainx runtime-report
./brainx knowledge-locate --query "deployment checklist"

Plugin Config

Start conservative:

{
  "plugins": {
    "entries": {
      "brainx": {
        "enabled": true,
        "config": {
          "wikiDigest": true,
          "jitRecall": false,
          "workingMemory": false,
          "toolAdvisories": false,
          "captureToolFailures": false,
          "bootstrapMode": "off",
          "captureOutboundMode": "off"
        }
      }
    }
  }
}

Enable JIT recall, working memory, advisories, and write-path capture only after privacy and retention rules are clear for the target installation.

Feature Groups

• Persistent vector memory with PostgreSQL + pgvector
• Semantic search and contextual injection
• Intelligent memory classification
• Hot/warm/cold lifecycle management
• Deduplication, contradiction detection, and supersede chains
• PII scrubbing and sensitivity controls
• Verification states: verified, hypothesis, changelog, obsolete
• Quality scoring and feedback loops
• Memory distillation and fact extraction
• Knowledge vault import, sync, and location
• Cross-agent learning with trust gates
• Context packs and wiki digest
• Runtime working memory and JIT recall
• Pre-action advisories
• Conservative tool-failure capture
• Doctor, auto-fix, metrics, and runtime reports
• Backup and restore scripts
• Eval dataset generation and retrieval tests

Public Distribution Rules

• Do not publish .env, database dumps, backups, runtime memory, tool-failure payloads, private knowledge vaults, or local workspace state.
• Do not hardcode private agent names, host paths, or user-specific operating context.
• Keep examples generic and portable.
• Keep secrets in environment variables only.