Backend Auth & Model Map
Last verified: 2026-05-04 (second live probe this session)
Status icons: ✓ working · ⚠ partial / pending · ✗ not built · 🆕 new since last map · 🔧 in progress
Backend Auth Audit
Backend Henry’s intent Current state (verified 2026-05-04) Gap Anthropic Max #1 Plan-limits, flat-rate :18903 anthropic-max-router ✓ WORKING — 8/8 tests passed Monitor “extra usage” credit balance teamsteph@betterfiles.com OAuth account for Max #1 ✓ Authenticated, bound to :18903 This is the Max plan account all 40+ agents use via LOCAL-SHORTCUT claude-max-api-proxy :18910 Alternative max proxy 🆕 ✓ RUNNING 127.0.0.1:18910 — hot standby Not wired to portkey-proxy; kept as CCR Phase 0 candidate Anthropic Max #2 (henry@reri.co ) Second Max plan :18911 ✗ not built — no listener on 18911 Mirror Phase 0 with henry@reri.co OAuth Codex (ChatGPT plan) OAuth via codex login ✓ LOGGED IN — auth_mode: chatgpt, henryhillre@gmail.com, refreshed 2026-05-04T14:30Z. ⚠ plan_type: plus (not Pro) — rate limits may be lower than expected Wire Codex subprocess into CCR Phase 2; confirm Plus vs Pro limits Kimi (kimi.com) OAuth via kimi login ✓ LOGGED IN — ~/.kimi/credentials/kimi-code.json populated: access_token (900s TTL) + refresh_token (30d). scope: kimi-code. Telemetry failures in logs are network-only, not auth. Wire Kimi subprocess into CCR Phase 3 DeepSeek API + OAuth ✓ DEEPSEEK_API_KEY + PORTKEY_DEEPSEEK_DIRECT_VK in master.env + OAuth set up Both API and OAuth paths available Gemma-4 (31B dense) OpenRouter via Portkey 🔧 IN PROGRESS — pc-opencl-2f97e0 config exists, model IDs mapped Verify OpenRouter VK has Gemma-4-31b access Gemma-4 (26B MoE) OpenRouter via Portkey 🔧 IN PROGRESS — pc-opencl-1373ad config exists, model IDs mapped Verify OpenRouter VK has Gemma-4-26b access Voyage-4 (embeddings) API VOYAGE_API_KEY + Portkey pc-opencl-22d508✓ working Perplexity Search/Sonar 🆕 PERPLEXITY_API_KEY in master.env + 4 Portkey configs live ⚠ MCP perplexity server connected; Phase 6 5-layer integration pending OpenAI (raw) API OPENAI_API_KEY in master.env✓ available — not in Portkey (direct) OpenRouter API fallback OPENROUTER_API_KEY in master.env✓ working fallback for Gemma/cloud models Gemini API GEMINI_API_KEY in master.env✓ available — not yet in Portkey config Groq API GROQ_API_KEY in master.env✓ available — not yet in Portkey config Ollama AWS Mac Cloud LLM host SSH timeout (100.123.248.46) — EC2 impaired ✗ unreachable — fix EC2 health first Ollama Mac Ultra Home M1 Ultra local LLMs OLLAMA_BRIDGE_HMAC in master.env; runs locally ⚠ not reachable from VPS — needs HMAC bridge or SSH key fix Moonshot direct API Superseded by Kimi OAuth MOONSHOT_API_KEY + KIMI_CODE_API_KEY in master.env — stale⚠ archive after Kimi OAuth verified in Phase 3
Portkey Configs (21 model entries, live tier-config.json, verified 2026-05-04)
All 21 entries audited against live tier-config.json. defaultConfig: pc-opencl-bd87ab.
Config slug Bound model ID(s) Provider VK / Route Status pc-opus-f-be2d19claude-opus-4-7 · claude-opus-4-6 · claude-opus-4primary anthropic-direc-40d4a5 + OpenRouter fallback on 400/401/429 ✓ live pc-sonnet-df3a4cclaude-sonnet-4-6 · claude-sonnet-4same fallback chain ✓ live pc-haiku-5c4a73claude-haiku-4-5-20251001 · claude-haiku-4-5same fallback chain ✓ live pc-opencl-22d508voyage-4 · voyage-3Voyage AI embeddings ✓ live pc-opencl-2f97e0gemma-4-31b-it · google/gemma-4-31b-itOpenRouter Gemma 31B dense 🔧 in progress pc-opencl-1373adgemma-4-26b-a4b-it · google/gemma-4-26b-a4b-itOpenRouter Gemma 26B MoE 🔧 in progress pc-opencl-bd87abdefault fallback API-tier anthropic-direc-40d4a5 ✓ live pc-perple-ee827asonar · sonar-pro · sonar-reasoning-pro · sonar-deep-research · perplexity/sonar · perplexity/sonar-pro · perplexity/sonar-reasoning-pro · perplexity/sonar-deep-researchPerplexity API ✓ live pc-opencl-98748edavid-socal agent only OpenRouter (Gemma/mixed) ✓ live (missing) Codex / Kimi / DeepSeek-direct / Gemini / Groq / cloud models below Not yet in Portkey ✗ Phase 4 CCR wires these
Postgres governance (CHOKEPOINT-2): Each Portkey config change must write an infra_config_changes row with change_type, target_system, after_state, reason, approved_by, applied_via before going live.
All Models (complete)
Model ID Tier Context Routing claude-opus-4-7Strategic 200k LOCAL-SHORTCUT → :18903 → claude.ai plan claude-opus-4-6 🆕Strategic 200k confirmed valid; same route claude-sonnet-4-6Operations 200k LOCAL-SHORTCUT → :18903 claude-haiku-4-5-20251001Automation 200k LOCAL-SHORTCUT → :18903 Anthropic Batches API Bulk background — Phase 4 CCR — high-volume async jobs
Max plan account: teamsteph@betterfiles.com — all 40+ agents tagged tier: max-plan in agentConfigs route here via LOCAL-SHORTCUT in proxy.js, bypassing Portkey cloud. henryRERI stays on interactive VS Code session.
models.json name vs id discrepancy: aurora and acquisitions agents have "id": "claude-opus-4-7" but "name": "claude-opus-4-6". The id field is what gets sent to Anthropic. Standardization pending.
Cloud Models — LLM Build (via OpenRouter / direct APIs)
Model ID Provider route Context Strengths CCR priority deepseek/deepseek-v3-2025OpenRouter 128k Code, reasoning, cheap P1 moonshotai/kimi-k2.5Kimi OAuth → kimi-cli 128k Long-context, Chinese P1 moonshotai/kimi-k2-turboKimi OAuth → kimi-cli 32k Fast Kimi variant P2 meta-llama/llama-4-maverick-17b-128e-instruct-fp8OpenRouter 128k Fast MoE, cheap P1 qwen/qwen3.5-397b-a17b-instruct-fp8OpenRouter 32k Massive dense, best-in-class reason P2 minimax/minimax-m2.5OpenRouter 1M Ultra-long context P3 x-ai/grok-4-fastOpenRouter 128k Speed, multimodal P2 google/gemini-3-flash-previewGemini API direct 1M Flash speed, multimodal P1 google/gemini-3-pro-previewGemini API direct 1M Pro quality P2 groq/llama-3.3-70b-versatileGroq API direct 128k Ultra-fast inference P1 deepseek/deepseek-r1OpenRouter + OAuth 64k Open-source reasoning P2 gemma-4-31b-itOpenRouter → pc-opencl-2f97e0 128k Portkey-wired dense 🔧 gemma-4-26b-a4b-itOpenRouter → pc-opencl-1373ad 128k Portkey-wired MoE 🔧
These models are in the CCR routing table but not yet flowing through Portkey. CCR Phase 4 wires them with per-model intent rules.
Local Ollama Models (M1 Ultra — not yet reachable from VPS)
Model Variant Context Role in CCR qwen3.5:122b-instruct-q4kmctx32k 32k default routing target qwen3.5:122b-instruct-q4kmctx64k 64k short doc analysis qwen3.5:122b-instruct-q4kmctx96k 96k medium context qwen3.5:122b-instruct-q4kmctx128k 128k long context qwen3.5:122b-instruct-q4kmctx192k 192k extended context qwen3.5:122b-instruct-q4kmctx262k 262k max local context qwen3.6:35b-a3b-instruct-q4km— 32k background tasks (cheap) qwen3:235b-a22b-thinking-q3km— 32k heavy local reasoning gemma4:26bMoE 128k light/fast local (preferred over 31b) gemma4:31bdense 128k quality local (⚠ low-priority — dominated by 26b MoE, 3-15× slower) deepseek-r1:32b— 32k math/reasoning only (⚠ lane:math-only — Ollama bug #8517 tool-reject)
Routing rule (CCR): default→qwen3.5:122b-ctx32k · think→122b-ctx96k (→235B) · longContext→122b-ctx262k · background→qwen3.6:35b · fallback→Portkey-Sonnet
Blocker: M1 Ultra SSH (Permission denied publickey) + EC2 impaired. Ollama install deferred until bridge fixed.
Agent Tier Map (44 active + 5 retired)
Tier 1 — Strategic (Opus) — 2 agents
Max-plan LOCAL-SHORTCUT · teamsteph@betterfiles.com account
Agent Role Discord aurora Chief AI Ops, Discord commander, primary builder aurora-build solara Strategic orchestration, system governance solara-build
Tier 2 — Operations (Sonnet) — 13 agents
Max-plan LOCAL-SHORTCUT · same teamsteph account
Agent Role Key Services main Central orchestration, 80%+ of cron jobs sm-token-refresh.serviceatlas Architecture, InvestorLift ops, infrastructure property-view-watcher.service, investorlift-gateway.serviceacquisitions Acquisitions engine, inbound lead handling — betterfiles Document management, TC coordination — dispo Disposition management, buyer outreach — bizintel Business intelligence, market analysis — reri RERI brand operations — ryan-norcal Sales — Northern California — david-socal Sales — Southern California Gemma route via pc-opencl-98748e (only agent NOT on max-plan LOCAL-SHORTCUT) hhill-socal Founder/executive ops — auctionscout Auction intelligence — prediction-trader Prediction market + crypto trading prediction-ingestor.servicebetterbets Multi-sport betting orchestrator —
Tier 3 — Automation (Haiku) — 18 agents
Max-plan LOCAL-SHORTCUT · same teamsteph account
Cluster Agents General automation research · ops · mktg · webops · social Real estate / finance backfill · bettershowings · betterhomeoffers · bettercredit · betterarbitrage · tradelines · crowdfunding Fund calls (parent + 3) fund-calls → angel-calls · h2-calls · h3-calls InvestorLift (parent + 3) investorlift → investorlift-angel · investorlift-h2 · investorlift-h3 Inactive bettertrading · tablegames
Special / Mixed-Model — 3 agents
Agent Primary Model Notes kimi moonshot-v1-128k / Kimi OAuthLong-context, Chinese-language tasks chatgpt gpt-4oGPT-4o bridge anthropic config-only Not a real agent — provider entry only
Retired / Archived — 5 (do not delete per feedback_archive_not_delete.md)
stefania · stefania-asst · default · betteracq (→ acquisitions) · dealcmd (→ dispo)
Full Pipeline Visual Map
┌──────────────────────────────────────────────────────────────────────────────┐
│ OPENCLAW REQUEST → DISPATCH PIPELINE │
└──────────────────────────────────────────────────────────────────────────────┘
44 AGENTS (gateway-embedded)
Aurora · Solara · Atlas · Acq · Dispo · BetterFiles · Main · 38 more
│
▼
┌──────────────────────────────────────────┐
│ openclaw-gateway :18789 (Go binary) │ ✓ running
│ Tags x-openclaw-agent header │
└─────────────────────┬────────────────────┘
▼
┌──────────────────────────────────────────┐
│ portkey-proxy :18900 (Node.js) │ ✓ running
│ - tier-config.json model→config map │
│ - 21 model entries (9 configs) │
│ - LOCAL-SHORTCUT → :18903 for max-plan │
│ agents (teamsteph@betterfiles.com) │
│ - Portkey cloud for fallback/API-tier │
│ - SIGHUP hot-reload (no restart needed) │
└─────────────────────┬────────────────────┘
▼
╔══════════════════════════════════════════════════════════════════════╗
║ CCR DISPATCHER :18902 (TO BUILD — Phase 4) ║ ✗ not built
║ claude-code-router (musistudio) ║
║ - intent classifier → route per-call ║
║ - per-account inflight tracking ║
║ - failover ladder per backend ║
║ - backend health probes ║
║ - CHOKEPOINT-1: writes tool_calls row to Postgres before return ║
║ - CHOKEPOINT-2: config changes write infra_config_changes first ║
║ - sends Langfuse trace per call ║
╚══╤════╤═══════╤═══════╤══════╤══════╤════════╤══════╤═══════╤══════╝
│ │ │ │ │ │ │ │ │
┌─────┘ │ ┌────┘ ┌───┘ ┌───┘ ┌───┘ ┌────┘ ┌───┘ ┌───┘
▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼
┌──────┐ ┌──────┐ ┌────┐ ┌────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌────┐ ┌──────┐
│Anth │ │Anth │ │Cod │ │Kimi│ │Deep │ │Gemini│ │Groq │ │Perp│ │Open │
│Max#1 │ │Max#2 │ │ex │ │ │ │Seek │ │API │ │API │ │lex │ │Router│
│:18903│ │:18911│ │ │ │ │ │ │ │ │ │ │ │ity │ │ │
│teamst│ │henry │ │ │ │ │ │ │ │ │ │ │ │ │ │ │
│ph ✓ │ │✗ not │ │✓ │ │✓ │ │✓ API │ │✓ key │ │✓ key │ │✓ MCP│ │✓ key│
│ │ │built │ │logd│ │logd│ │+OAuth│ │in env│ │in env│ │+key│ │in env│
└──┬───┘ └──┬───┘ └─┬──┘ └─┬──┘ └──┬───┘ └──┬───┘ └──┬───┘ └─┬──┘ └──┬───┘
│ │ │ │ │ │ │ │ │
▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼
┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌────────┐ ┌──────┐ ┌──────┐ ┌──────┐
│anth- │ │anth- │ │codex │ │kimi- │ │deepsk │ │gemini│ │llama │ │sonar │
│max- │ │max- │ │CLI │ │CLI │ │v3.2/r1 │ │3-fls/│ │3.3- │ │/pro/ │
│router│ │api- │ │subprc│ │subprc│ │direct │ │pro │ │70b │ │rsn/ │
│:18903│ │proxy │ │ │ │ │ │VK in │ │API │ │Groq │ │deep │
│✓LIVE │ │:18911│ │ │ │ │ │Portkey │ │ │ │fast │ │rsrch │
│(teamst│ │✗NOT │ │chatg │ │kimi │ │ │ │ │ │ │ │ │
│efan) │ │BUILT │ │pt.com│ │.com │ │ │ │ │ │ │ │ │
└──────┘ └──────┘ └──────┘ └──────┘ └────────┘ └──────┘ └──────┘ └──────┘
┌─────────────────────────────────────────────────────────────────┐
│ claude-max-api-proxy :18910 🆕 ✓ RUNNING (hot standby) │
│ Not wired to portkey-proxy — available for CCR Phase 0 cutover │
└─────────────────────────────────────────────────────────────────┘
══════════════════ ADDITIONAL BACKENDS (retained — CCR Phase 4) ═══════════════
Cloud LLMs via OpenRouter (single API key, all models below):
┌──────────────────┐ ┌──────────────────┐ ┌──────────────────┐
│ deepseek-v3.2 │ │ llama-4-maverick │ │ qwen3.5-397b │
│ (code/reason) │ │ 17B-128E MoE │ │ dense FP8 │
│ P1 │ │ P1 │ │ P2 │
└──────────────────┘ └──────────────────┘ └──────────────────┘
┌──────────────────┐ ┌──────────────────┐ ┌──────────────────┐
│ grok-4-fast │ │ minimax-m2.5 │ │ deepseek-r1 │
│ xAI fast │ │ 1M context │ │ OSS reasoning │
│ P2 │ │ P3 │ │ P2 │
└──────────────────┘ └──────────────────┘ └──────────────────┘
Kimi-direct via OAuth (kimi-cli subprocess):
┌──────────────────┐ ┌──────────────────┐
│ kimi-k2.5 │ │ kimi-k2-turbo │
│ 128k context │ │ 32k fast variant │
│ P1 │ │ P2 │
└──────────────────┘ └──────────────────┘
Gemini direct API (GEMINI_API_KEY in master.env):
┌──────────────────┐ ┌──────────────────┐
│gemini-3-flash │ │gemini-3-pro │
│1M ctx, fast, P1 │ │1M ctx, quality P2│
└──────────────────┘ └──────────────────┘
Groq (ultra-fast inference): Codex (ChatGPT plan OAuth):
┌──────────────────┐ ┌──────────────────┐
│ llama-3.3-70b │ │ ChatGPT plan │
│ ~800 tok/s P1 │ │ ✓ logged in │
└──────────────────┘ └──────────────────┘
Ollama (M1 Ultra local — pending bridge fix):
┌──────────────────┐ ┌──────────────────┐ ┌──────────────────┐
│ qwen3.5:122b │ │ qwen3:235b-think │ │ qwen3.6:35b-a3b │
│ 6 ctx variants │ │ heavy reasoning │ │ background tasks │
│ default target │ │ q3km quant │ │ cheap │
└──────────────────┘ └──────────────────┘ └──────────────────┘
┌──────────────────┐ ┌──────────────────┐ ┌──────────────────┐
│ gemma4:26b MoE │ │ gemma4:31b dense │ │ deepseek-r1:32b │
│ light/fast │ │ quality local │ │ local reasoning │
└──────────────────┘ └──────────────────┘ └──────────────────┘
Blocker: SSH key + EC2 impaired (100.123.248.46). OLLAMA_BRIDGE_HMAC in master.env ✓
┌──────────────────────────────────────────────────────────────────────────────┐
│ OBSERVABILITY + PERSISTENCE FAN-OUT │
│ (every CCR-routed call writes to ALL of the below — CHOKEPOINT-1) │
└──────────────────────────────────────────────────────────────────────────────┘
│
┌─────────────────────────┼─────────────────────────┐
▼ ▼ ▼
┌──────────────────┐ ┌──────────────────┐ ┌────────────────────────┐
│ PORTKEY (cloud) │ │ LANGFUSE :18840 │ │ SUPABASE CCP │
│ ✓ live │ │ ✓ env vars live │ │ svueekfvfrvhylxygktb │
│ │ │ OSIL Langfuse │ │ ✓ live, 373 tables │
│ 9 Portkey cfgs │ │ Docker :18840 │ │ │
│ (7 Anthropic + │ │ Worker :18841 │ │ tool_calls table │
│ 2 Gemma 🔧 + │ │ LANGFUSE_HOST │ │ CHOKEPOINT-1: every │
│ 4 Perplexity✓ +│ │ LANGFUSE_PUBLIC │ │ LLM call → row here │
│ 1 default + │ │ LANGFUSE_SECRET │ │ BEFORE returning │
│ 1 OpenRouter) │ │ LANGFUSE_ROTATE │ │ │
│ │ │ │ │ infra_config_changes │
│ Pro plan $49/mo │ │ trace per call │ │ CHOKEPOINT-2: config │
│ logging, retry │ │ phoenix-arize │ │ changes written FIRST │
│ cache observe │ │ eval hooks (T10)│ │ │
└──────────────────┘ └──────────────────┘ │ omni_events │
│ 9.85 GB HNSW idx │
│ tool_calls_fallback: │
│ /tmp/openclaw/ │
│ tool-calls-fallback │
│ .jsonl (if PG down) │
└───────────┬────────────┘
▼
┌────────────────────────────────────────────────────────────┐
│ EMBEDDING + HYBRID SEARCH │
└────────────────────────────────────────────────────────────┘
│
┌────────────────────────────────────┼────────────────────────────────┐
▼ ▼ ▼
┌────────────────────┐ ┌──────────────────────────────┐ ┌───────────────────┐
│ EMBEDDING PIPELINE│ │ HYBRID SEARCH │ │ PER-AGENT SQLite │
│ │ │ │ │ │
│ Voyage-4 primary │ │ 60 vector cols total │ │ 44 DBs in │
│ ✓ via Portkey │ │ 52 HNSW indexes (halfvec) │ │ memory/*.sqlite │
│ pc-opencl-22d508 │ │ halfvec migration in-flight │ │ ~5,568 chunks │
│ │ │ │ │ vec0 + FTS5 │
│ Voyage-3 fallback │ │ 7 hybrid_search_* RPCs │ │ 70% vector / │
│ pc-opencl-22d508 │ │ ✗ UNWIRED — 0 calls 5 mo │ │ 30% keyword │
│ │ │ │ │ │
│ Perplexity embed │ │ Coverage today: │ │ SEMANTIC MCP: │
│ ⚠ deferred │ │ salesmsg_inbox 30.9% │ │ 8 semantic_ │
│ (Phase 6 sub.) │ │ omni_events 43.4% │ │ search_* tools │
│ │ │ thread_msgs 42.1% │ │ ✗ not built │
│ embed-if-text │ │ knowledge_docs 0.0% ✗ │ │ Phase 8 sub.plan │
│ ✗ not scheduled │ │ │ │ │
│ (Phase 5 sub.) │ │ mem0 / Honcho patterns │ │ acquisitions. │
│ │ │ ⚠ audit pending (OSIL T4) │ │ sqlite only 1 │
└────────────────────┘ └──────────────────────────────┘ │ chunk ⚠ Phase │
│ 7.5 investigation│
└───────────────────┘
┌──────────────────────────────────────────────────────────────────────────────┐
│ AUDIT + DRIFT DETECTION │
└──────────────────────────────────────────────────────────────────────────────┘
┌──────────────────┐ ┌───────────────────┐ ┌──────────────────┐ ┌──────────────────┐
│ cost-monitor.js │ │ models-guardian.sh │ │ security-audit- │ │ tool-calls- │
│ ⏸ PAUSED │ │ ✓ running │ │ funnel │ │ health-check.js │
│ │ │ │ │ ✓ Mon 06:00 PT │ │ ✓ every 5 min │
│ TODO: add CCR │ │ guards models.json │ │ │ │ CHOKEPOINT-1: │
│ fallback header │ │ from drift on │ │ Detects: │ │ compares Portkey │
│ check before │ │ gateway restart │ │ - unregistered │ │ call count vs │
│ re-enable │ │ │ │ webhooks │ │ tool_calls rows │
│ │ │ TODO: rewrite for │ │ - sig fail >5% │ │ >10% delta → │
│ │ │ CCR semantics │ │ - stale WAF rules│ │ Discord #ops │
└──────────────────┘ └───────────────────┘ └──────────────────┘ └──────────────────┘
┌───────────────────────────────────────────────────────────────────────────────────┐
│ claude-orphan-reaper.js ✓ running (5-min timer) │
│ 🆕 FIXED 2026-05-04: hasActivePipe() guard — checks /proc/PID/fd/0 for │
│ socket/pipe link. Skips reaping if stdin is active VS Code socket. │
│ Was: killing live VS Code sessions (root cause of exit code 143). Now fixed. │
└───────────────────────────────────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────────────────────────┐
│ MCP TOOLS (live in .mcp.json — 21 servers) │
└──────────────────────────────────────────────────────────────────────────────┘
Core 8 (always loaded):
┌────────────────────┐ ┌────────────────────┐ ┌────────────────────┐
│ openclaw-tools │ │ aurora-orchestrator │ │ context7 │
│ ✓ 11 tools: │ │ ✓ 20 tools: │ │ ✓ 2 tools: │
│ memory_search │ │ hubspot_lookup │ │ resolve-library-id│
│ memory_save │ │ hubspot_deal_* │ │ query-docs │
│ workspace_query │ │ semantic_search │ │ (library docs) │
│ run_script │ │ openphone_thread │ └────────────────────┘
│ portkey_status │ │ schedule_followup │ ┌────────────────────┐
│ agent_status │ │ supabase_query │ │ playwright │
│ gateway_logs │ │ memory_search │ │ ✓ headless browser │
│ supabase_query │ │ omni_contact_ │ │ DOM/screenshot │
│ hubspot_lookup │ │ timeline │ │ JS execution │
│ memory_health │ │ reply_to_imessage │ └────────────────────┘
│ stack_query │ │ ...more │
└────────────────────┘ └────────────────────┘
┌────────────────────┐ ┌────────────────────┐ ┌────────────────────┐
│ shortio │ │ 21st-magic │ │ figma │
│ ✓ 11 tools: │ │ ✓ 4 tools: │ │ ✓ design files │
│ create-short-link │ │ component_builder │ │ (read-only) │
│ edit-link │ │ component_refine │ └────────────────────┘
│ link-statistics │ │ component_inspire │
│ domain-statistics │ │ logo_search │
└────────────────────┘ └────────────────────┘
Additional active (partial load):
┌────────────────────┐ ┌────────────────────┐ ┌────────────────────┐
│ perplexity │ │ firecrawl │ │ sequential-thinking│
│ ✓ 4 tools: │ │ ✓ tools: │ │ ✓ 1 tool: │
│ perplexity_ask │ │ firecrawl_scrape │ │ sequentialthink │
│ perplexity_reason │ │ firecrawl_search │ │ (chain-of-thought)│
│ perplexity_search │ │ firecrawl_crawl │ └────────────────────┘
│ perplexity_rsrch │ │ firecrawl_extract │
└────────────────────┘ └────────────────────┘
┌────────────────────┐ ┌────────────────────┐ ┌────────────────────┐
│ memory (graph) │ │ github │ │ supabase │
│ ✓ 10 tools: │ │ ✓ 40+ tools: │ │ ✓ tools: │
│ create_entities │ │ search_code │ │ execute_sql │
│ create_relations │ │ list_issues │ │ apply_migration │
│ read_graph │ │ create_pr │ │ get_logs │
│ search_nodes │ │ get_commit │ │ list_tables │
│ add_observations │ │ push_files │ │ ...more │
└────────────────────┘ └────────────────────┘ └────────────────────┘
┌────────────────────┐ ┌────────────────────┐ ┌────────────────────┐
│ dune │ │ airtable │ │ notion │
│ ✓ (HTTP mode) │ │ ✓ tools: │ │ ✓ tools: │
│ on-chain queries │ │ list_bases │ │ post-page │
│ dashboard create │ │ list_records │ │ retrieve-database │
│ query execute │ │ create_record │ │ query-datasource │
└────────────────────┘ └────────────────────┘ └────────────────────┘
┌────────────────────┐ ┌────────────────────┐ ┌────────────────────┐
│ discord │ │ slack │ │ hubspot (direct) │
│ ✓ 18 tools: │ │ ✗ PLACEHOLDER │ │ ✗ token stub │
│ send_message │ │ token not set │ │ (uses aurora-orc │
│ get_history │ │ │ │ hubspot_lookup │
│ add_reaction │ └────────────────────┘ │ instead) │
│ edit_message │ ┌────────────────────┐ └────────────────────┘
└────────────────────┘ │ stripe │ ┌────────────────────┐
│ ✗ PLACEHOLDER │ │ google-workspace │
│ token not set │ │ ✗ wrong creds file │
└────────────────────┘ └────────────────────┘
┌────────────────────┐
│ postgres-diag │
│ ✗ env var not │
│ expanded (known) │
└────────────────────┘
MCP servers MISSING from .mcp.json (not yet installed):
✗ Perplexity MCP full 5-layer — Phase 6 of Perplexity plan (perplexity server IS
connected but full 5-layer integration pending Q1-Q7)
✗ semantic_search_* RPCs — Phase 8 of substrate plan
✗ CCR MCP — N/A (CCR is a backend service, not MCP)
Governance Gate Compliance
Gate Rule Status CHOKEPOINT-1 Every LLM call writes tool_calls row before returning ⚠ CCR must implement when built. tool-calls-health-check.js monitors drift CHOKEPOINT-2 Config changes write infra_config_changes BEFORE going live ✓ Required for every Portkey config edit CHOKEPOINT-3 Schema migrations via workspace/migrations/ files only ✓ No raw ALTER TABLE from chat G-NO-PLAINTEXT-CREDS All credentials via op://Aurora/<item>/<field> ⚠ master.env is chmod 600 but still plaintext — op:// migration in progress G-SERVICE-PRE-START-DOC New services in CLAUDE.md port map + ARCHITECTURE.md before first start ✓ :18903/:18910 documented G-FAILED-SERVICE-MTTR Failed services fixed/disabled/archived within 24h Monitor via daily cron G-MODEL-ROUTING-AT-EXEC Execution mode dispatches Sonnet first via Agent tool ✓ Active this session G-BLOCKER-SURFACING Open blockers mirrored to SESSION-AUDIT.md same session ✓ See Open Decisions below
CCR Migration Plan Status (12 phases)
Phase Description Status 0 Stand up claude-max-api-proxy@teamsteph :18910, smoke test 🆕 :18910 RUNNING — :18903 also working. Keep both (Q1) 1 Add henry@reri.co Max #2 on :18911 ✗ not built 2 Codex auth — wire Codex CLI subprocess into CCR ✓ Codex logged in — CCR wiring pending Phase 4 3 Kimi auth — wire kimi-cli subprocess into CCR ✓ Kimi logged in — CCR wiring pending Phase 4 4 CCR install + base config (:18902, intent classifier, routing rules) ✗ not built 5 Wire instrumentation fan-out (Postgres tool_calls, Langfuse) ✗ not built 6 Perplexity wire-in (depends on Q1-Q7 answers) ⚠ partial — API key + Portkey live; MCP connected; full 5-layer pending 7 Substrate touchpoints (coordinate with substrate plan) ✗ not built 8 Cut proxy.js LOCAL-SHORTCUT → CCR (:18902) ✗ not built 9 Update tier-config.json semantics (route field replacing max-plan tag) ✗ not built 10 Audit fix + un-pause cost-monitor ✗ paused 11 Decommission anthropic-max-router, archive stale env vars ✗ pending Phase 0 decision 12 Ollama install (AWS Mac) + Mac Ultra reachability ✗ EC2 impaired, SSH timeout
Open Decisions
# Question Recommendation Q1 :18903 (anthropic-max-router) now working — keep alongside :18910 or cut over to claude-max-api-proxy? Keep :18903 as primary (it works), wire :18910 as hot-standby Q2 New plan ccr-server-side-dispatcher OR amend misty-valley? New plan — different scope Q3 ChatGPT account for Codex OAuth Phase 2 wiring? Whichever holds the $200 ChatGPT plan Q4 Archive MOONSHOT_API_KEY + KIMI_CODE_API_KEY after Phase 3? Yes, after Kimi OAuth subprocess verified Q5 Ollama AWS Mac — block on EC2 health fix first? Yes — resolve EC2 impaired status first Q6 Perplexity Q1-Q7 in perplexity plan need answers before Phase 6 Answer next session Q7 Gemma-4 OpenRouter VK access — verify 31B + 26B models accessible? Smoke-test: curl OpenRouter with model ID, check 200 Q8 Gemini + Groq — add to Portkey configs or route direct via CCR? Direct via CCR intent router (no Portkey overhead for fast models) Q9 Anthropic Batches API — which agents should use it? research/backfill/social for bulk background jobs (Phase 4 CCR rule)
What Changed Since Last Map Version
Change Detail ✓ Codex logged in Updated from ⚠ API-key mode ✓ Kimi logged in Updated from ⚠ empty credentials ✓ DeepSeek OAuth Both API + OAuth available 🔧 Gemma-4 added gemma-4-31b-it + gemma-4-26b-a4b-it in tier-config (in progress) ✓ teamsteph@betterfiles.com labeled Max plan account correctly labeled throughout 📋 All cloud models added deepseek-v3.2, kimi-k2.5/turbo, maverick, qwen3.5-397b, minimax-m2.5, grok-4-fast, gemini-3-flash/pro, groq/llama-3.3-70b, deepseek-r1, Batches API 📋 All local Ollama models added qwen3.5:122b (6 ctx), qwen3.6:35b, qwen3:235b-thinking, gemma4:26b/31b, deepseek-r1:32b 📋 Additional backends inside CCR section Moved from footnote to inside pipeline diagram 📋 Portkey configs fully audited All 21 model entries verified against live tier-config.json 📋 Governance gates added CHOKEPOINT-1/2/3 + G-gates table added 📋 MCP tools expanded All 21 .mcp.json servers documented (was 8) — includes perplexity, firecrawl, sequential-thinking, memory graph, github, supabase, dune, airtable, notion, discord, slack✗, stripe✗, google-workspace✗, postgres-diag✗ ✓ :18903 working 8/8 tests passed (Sonnet/Opus4.7/Opus4.6/Haiku × direct + proxy) ✓ :18910 running claude-max-api-proxy now on 127.0.0.1:18910 (hot standby) 🆕 claude-opus-4-6 Confirmed valid Anthropic model. In tier-config. Works through max-plan route. 🆕 Orphan reaper fixed hasActivePipe() guard — no longer kills active VS Code sessions (exit 143 fix)