AI / OSIL Domain Hub

OSIL — OpenClaw Self-Improvement Layer — is the compound-learning substrate that makes the platform smarter without Henry writing new code. 56 mechanisms catalogued across 22 categories and 12 blind spots, organized into 14 sequential tiers from eval infrastructure (T1) through autonomous roadmap execution (T14). OSIL is the central domain for all AI self-improvement, model routing, evaluation pipeline, and agent evolution work running on the OpenClaw VPS and EC2 Mac Ultra.

Quick reference

Field	Value
Master plan	openclaw-self-improvement-layer-2026-05-03
Phase status	Phase 0 ✅ DONE · Phases 1-14 in sequence
Primary agent	_summary
LLM router	portkey (:18900)
Eval infra	Langfuse (:18840/:18841) · DeepEval · PromptFoo
Memory system	SQLite vec0+FTS5 · 44 DBs · 5,568 chunks
NemoClaw status	pre-operational — nemoclaw
LiveKit status	smoke-tested · awaiting Sandbox live test
10DLC blocking	⚠️ gates ALL SMS tiers — osil-twilio-10dlc-resubmission-2026-05-03
IL AI displacement	B13 track — osil-il-ai-replication-2026-05-03
OSIL master stub	_OSIL

OSIL 14-tier stack

Tier	Name	Status	Key output
T1	Eval infrastructure (Langfuse/DeepEval)	planned	eval DB, trace logging
T2	DSPy/GEPA auto-optimization	planned	prompt optimizer
T3	Reflexion self-critique loop	planned	critique agent
T4	Voyager-style skill discovery	planned	skill composer
T5	TextGrad gradient learning	planned	weight tuning
T6	OpenSpace diff engine	track-only	diff + quality schema
T7	Mem0/Honcho user memory	planned	cross-session context
T8	Letta OS-level agent substrate	planned	agent OS
T9	EvoAgentX evolution loop	planned	generational eval
T10	Langfuse + PromptFoo CI eval	planned	eval CI gate
T11	LangFuse session replay	planned	session audit
T12	Patronus adversarial testing	planned	red-team suite
T13	NemoClaw cred-proxy pattern	B1-B6 pending	L7 cred injection
T14	Autonomous roadmap execution	deferred	self-planning agent

Sub-projects

Sub-project	Status	Blocker
osil-twilio-10dlc-resubmission-2026-05-03	B14 pending B14a Henry signoff	gates ALL OSIL SMS tiers
osil-il-ai-replication-2026-05-03	B13 pending B13a	IL AI displacement, no subscription
nemoclaw-audit-2026-05-03	B1-B6 pending ratification	L7 cred-proxy pattern
LiveKit voice substrate	smoke-tested ✅ · Sandbox live test pending	B15 pending
OpenSpace patterns	track-only (3 patterns stolen)	not full integration

Infrastructure this domain uses

• anthropic — LLM brain (Opus/Sonnet/Haiku via :18900)
• portkey — CHOKEPOINT-1 router + tool_calls enforcement
• supabase — tool_calls table + infra_config_changes + eval DB
• hetzner — VPS compute (CCX43 → right-size pending)
• aws — EC2 Mac Ultra (IL scraping, OSIL pilot VM candidate)
• github — vault + skills repo
• discord — ops alerts + eval result channel
• nemoclaw — cred-proxy stub (pre-operational)

Knowledge base docs

• README — DSPy auto-optimization patterns
• README — GEPA prompt evolution
• README — Reflexion self-critique
• README — Voyager skill discovery
• README — Honcho user modeling
• README — Mem0 cross-session memory
• README — Letta agent OS
• README — TextGrad gradient tuning
• README — DeepEval eval framework
• README — PromptFoo CI eval
• README — TruLens tracing
• README — Langfuse eval + session replay
• README — Karpathy auto-research patterns
• README — AgentSkills composition
• README — EvoAgentX evolution
• README — OpenSpace diff engine (track-only)
• README — LiveKit voice substrate
• README — VAPI voice (deferred)
• README — Retell voice (deferred)
• README — Instructor structured output
• README — LangGraph workflow
• README — AutoGen multi-agent
• README — MetaGPT team simulation
• README — Arize Phoenix tracing
• README — Patronus adversarial eval
• README — HALO alignment framework

Cross-links

Agents that touch this

• _summary — primary brain + OSIL Phase 0 executor
• _summary — orchestration + plan governance
• _summary — cron dispatcher (80% bottleneck, rebalance candidate)

Skills that invoke this

• wave-execute — self-correcting wave framework (formalization pending)
• acquisitions-outreach — Stage 1 compliance gates (OSIL Phase 1.7 gates)
• acquisitions-followup — Stage 2-4 (gates blocked by 10DLC)

Plans that govern this

• openclaw-self-improvement-layer-2026-05-03 — master (14 tiers) · ACTIVE
• openclaw-fragmentation-fix-2026-05-01 — gate enforcement (Amendment §A1) · IN FLIGHT
• nemoclaw-audit-2026-05-03 — cred-proxy pattern · PENDING B1-B6
• osil-twilio-10dlc-resubmission-2026-05-03 — 10DLC re-filing · BLOCKING
• osil-il-ai-replication-2026-05-03 — IL AI displacement · B13 PENDING

Feedback rules

• feedback_chokepoint_principle — every LLM call writes tool_calls row
• feedback_dual_write_required — state mutation = Supabase CCP
• feedback_agent_identity_first — SOUL.md first before any agent change
• feedback_no_plaintext_creds — G-NO-PLAINTEXT-CREDS always
• feedback_model_routing_at_exec — G-MODEL-ROUTING-AT-EXEC
• feedback_model_routing_pattern — Opus plans · Sonnet executes · Haiku trivial

System maps

• vm-osil-overview — visual OSIL architecture
• agents-tier-structure — agent tier hierarchy
• cost-flow — LLM cost routing

Related domain hubs

• real-estate — primary revenue domain OSIL serves
• trading — BetterBets (Wave 8, deferred)
• crypto — trading infrastructure
• g-gates-network — cascade-failure gates
• cron-timer-registry — 56 timers incl OSIL eval crons
• mcp-registry — 22 MCP servers, 4 broken connections

OSIL cluster (cross-hub)

_OSIL · anthropic · portkey · supabase · nemoclaw · aws · g-gates-network · plan-governance

Open issues / TODOs

• B1-B6 NemoClaw ratification pending Henry — blocks cred-proxy pattern (Tier 13)
• B14a Henry signoff — unblocks all OSIL SMS tiers (Twilio 10DLC gating)
• B13a Henry signoff — unblocks IL AI displacement track
• B15 pending — LiveKit Sandbox live test
• Phase E.2 systemd timer arming — wikilink auto-suggester at 03:30 PT
• main agent ~80% cron bottleneck — rebalance candidate (Wave 3+)
• acquisitions.sqlite 1-chunk anomaly — Phase 7.5 investigation pending
• VPS Cascade Permanent Remediation — Phase 1+2 in flight

Recent activity

• 2026-05-04: hub created (W3-S3)
• 2026-05-03: OSIL Phase 0 completed; B1-B12+B16 ratified; LiveKit smoke-tested
• 2026-05-03: Wave 1+2 hub authoring shipped 50 hubs; OSIL sub-projects created