🔀 Context Pipeline
Multi-instance context processing with trust-based anonymisation
🟢 Running OpenClaw Sessions
📦 Config Presets
🤖 Model Overrides
🔗 Session Bindings
📦 UAML Context Pipeline (last assembly)
🎚️ UAML Context Allocation
Control how much of the model's context window UAML can use for memory injection.
🔒 Anonymisation
Controls what gets anonymised based on model trust level. Override per-category rules below.
🔧 MCP Tool Filtering
UAML filters MCP tools based on detected intent. Only relevant tools are recommended to the model.
🛡️ Anonymisation Monitor
Real-time view of anonymised entities in the current context. Verify that sensitive data is properly masked before reaching the model.
📁 Context Snapshots
Saved context payloads (last 10). Download for analysis or session recovery.
⏱️ End-to-End Timing
Full request lifecycle: message → context assembly → LLM → response.
📊 Processing History
Recent UAML context assembly records with timing and source breakdown.
⚙️ Configuration
Model Registry
🧠 Prompt Profiles
Optimised prompt strategies per model type. Based on empirical testing (40 experiments per model).
🖥️ Host Configurations
Model performance varies by host. Context windows may be auto-reduced on limited VRAM.
🎯 Task Routing
Recommended model + host for each task type. Primary and fallback options.
📝 System Prompt Templates
Per-model, per-task system prompt templates. Thinking models require structured prompts.
Test Pipeline Recall
Run a full pipeline recall to test trust-based anonymisation, budget allocation, and focus engine behaviour.
📈 Pipeline Telemetry
Per-stage timing across all connected agents (OpenClaw, Hermes, Claude Code, LLM proxy).
Sources: MCP _log_pipeline_telemetry (Sprint 0) + LLM proxy proxy:<stage> rows (Sprint 4).
Project filter respects §10 RBAC. Click a tool row to drill down per-session.