📊 v0.4.5 BYOK Live + W09 Verified Metrics — March 1, 2026

[creator35lwb-web]

v0.4.5 BYOK Live + W09 Verified Metrics — March 1, 2026

v0.4.5 Release: Bring Your Own Key (BYOK) Live

VerifiMind PEAS v0.4.5 introduces per-tool-call BYOK — users can now override the default Gemini provider with their own API keys on any tool call. This was triple-validated by three independent AI agents before release.

Key Features:

• Per-tool-call provider override via api_key + llm_provider parameters on all 4 core tools
• Auto-detect key format: gsk_ → Groq, sk-ant- → Anthropic, sk- → OpenAI/compatible
• Ephemeral security: keys are never stored, logged, or cached — used only for the single inference call
• Backward compatible: existing users see zero changes; BYOK is opt-in per call

Implementation: PR #55

---

W09 Verified Service Metrics (Feb 23 – Mar 1, 2026)

All metrics are verified from GCP Cloud Run production logs. Scrapers excluded. Owner traffic excluded. Conservative rounding applied.

Headline Numbers

Metric	Value	Source
Engagement Hours (All-Time)	1,729.5 hours	GCP logs, owner excluded
Value Confirmation Rate	94.6% (940/994 sessions)	Follow-up prompt analysis
MCP Integration Rate	77.6%	Programmatic traffic classification
Weekly Active Users	165	Unique user-agent fingerprints
MCP Tools	10 (4 Core + 6 Template)	Server tool registry
LLM Providers	7+ (Multi-Model + BYOK)	Provider factory

Traffic Classification (W09)

Category	Hours	Share
MCP Client (Tool Users)	1,341.7h	77.6%
API Integration	281.2h	16.3%
Human Browser	106.6h	6.2%
Scraper (excluded)	0h	0%

Adoption Trajectory (Cumulative Flying Hours)

Week	Period	Weekly Hours	Cumulative
W02	Jan 06–12	38.0h	38.0h
W03	Jan 13–19	115.3h	153.3h
W04	Jan 20–26	262.4h	415.7h
W05	Jan 27–Feb 02	309.6h	725.3h
W06	Feb 03–09	425.4h	1,150.7h
W07	Feb 10–16	370.0h	907.9h
W08	Feb 16–22	404.5h	1,312.5h
W09	Feb 23–Mar 1	417.0h	1,729.5h

Inference Quality

W02–W08 operated in mock mode due to a deprecated model endpoint. Starting v0.4.4 (Feb 27, 2026), all three Trinity agents return real AI inference with _overall_quality: "full". W09 shows 70% real inference — the transition is underway.

Client SDK Distribution

Client	Share
Node.js	65.3%
Python SDK	20.3%
Browser	8.5%
Claude/Anthropic	4.5%
Other	1.4%

---

Smithery Hosting Sunset — March 1, 2026

Today marks the Smithery free hosting sunset. VerifiMind PEAS is unaffected — our MCP server is self-hosted on GCP Cloud Run with a custom domain (verifimind.ysenseai.org). We never depended on Smithery hosting. The Smithery CLI and registry remain active for discovery.

---

Updated Roadmap

The full updated roadmap is available at ROADMAP.md.

Current priority chain: v0.4.5 BYOK Live ✅ → v0.5.0 Foundation (next) → v0.5.x Stability → v0.6.0 Agent Skills → v0.7.0 MCP App

---

Metrics methodology: All numbers are derived from GCP Cloud Run production logs. Owner traffic and bot/scraper traffic are excluded. Conservative rounding is applied. The Value Confirmation Rate measures the percentage of sessions where users send a follow-up prompt after receiving an initial response, indicating they found value in the interaction.

Live dashboard: verifimind.io/#metrics