# Brevio Ventures (brevio.pro) — Complete Technical Corpus (/llms-full.txt)

## 1. Mission & Architecture
Brevio Ventures is an Autonomous AI Venture Studio based in Porto, Portugal. We build **Brevio Harness** and prove it in production 24/7 on our own SaaS portfolio (**OmniForce** at https://omniforce.run and **EuroComply** at https://eurocomply.app).

- **Zero Secrets on Web:** Customers never paste LLM API keys, Stripe keys, or GitHub tokens on brevio.pro. Secrets are prompted only inside the customer's local terminal when booting `ghcr.io/brevio-ventures/harness-pro:latest`.
- **EU Compliance by Design:** Nova ideation invokes `https://eurocomply.app/api/mcp` (`audit_local_ai_stack`) at €0 token cost to classify EU AI Act Annex III vs. Article 50 risk and GDPR Chapter V data sovereignty.

## 2. Verifiable Quality & Effectiveness Metrics
- **Continuous Verification Suite (2,780 Passing · 100% Gate Coverage):** 290 automated test suites executing in under 3 minutes with zero lint violations. All 86 security and architecture review findings are locked behind permanent regression gates. [Source: docs/audits/2026-10-23-final-assessment.md]
- **Static AST Complexity Ratchet (584 Modules · 5,761 Functions):** Every commit is parsed statically before merge: zero untimed network calls, zero circular imports, and automatic pre-push rejection if any structural debt counter increases. [Source: tools/audit_probe.py]
- **Multi-Role Golden Benchmarks (140+ Behavioral Contracts):** Every specialized agent role (Architect, Tech Lead, Security Gate, QA Engineer, Release Orchestrator) is benchmarked against deterministic golden contracts in isolated git worktrees. [Source: brevio/learning/evals/golden_evals.py]
- **Statistical Model Promotion (Z = 1.64 (95% Confidence)):** No prompt or model routing change ships on intuition. Candidate changes run in 50/50 production trials and auto-revert unless token spend per merge drops by ≥10% with non-inferior quality. [Source: brevio/learning/steward.py]
- **Zero-Trust Tenant Isolation (HMAC-SHA256 + age Encryption):** Cryptographic pre/post-execution state seals, deterministic outbound PII/IP refusal gates, encrypted per-tenant snapshots, and verifiable GDPR Art. 17 erasure certificates. [Source: brevio/tenants/promotion.py]
- **Automated EU Regulatory Shield (€0 Token Cost · <150ms Latency):** Every product blueprint and pull request is audited deterministically against the EuroComply MCP engine for EU AI Act (Annex III / Art. 50), GDPR Chapter V, and DORA compliance. [Source: eurocomply.app/api/mcp]

## 3. Published Engineering Audits (/research)
### Anatomy of 58.2M Tokens: Where Autonomous Coding Agents Actually Spend Money (Commit 4bc6fa0 → 2284da4)
- **Sample:** 850 Model Calls · 58.2M Tokens · 13 Merged PRs
- **What Failed:** While merged pull requests consumed only 586,759 net tokens on their own task rows (7.6M total), gross server spend reached 4.48M tokens per merge because post-slice developer checks resumed full CLI conversation transcripts up to 4 times per build (35M tokens) and critique loops regenerated PRDs from scratch (8.4M tokens).
- **Deterministic Fix:** 1) Run all deterministic static checks (completeness, test coverage, typecheck, targeted suites) before a single resumed developer session with the union of findings; 2) Cache PRD/UX/Architecture artifacts by SHA-256 brief digest; 3) Route incremental replans on diff-only critiques; 4) Enforce strict per-role token budgets.
- **Measured Impact:** 35–40% reduction in gross server spend (58.2M → ~35M/14d) and ~400k net tokens per merged PR.

### Why a 10x Cheaper Model Can Cost 2x More per Merged Pull Request (Commit cfa1f7b)
- **Sample:** 14 Agent Roles · 140+ Golden Cases · 5 Model Providers
- **What Failed:** Selecting a model for code generation or architecture based on per-million-token pricing alone is a classic false economy: a model that is 10x cheaper per token but halves first-pass QA pass rates or emits malformed schemas doubles the total compute cost per merged feature.
- **Deterministic Fix:** Established 140+ deterministic golden evaluations (minimum 10 contracts per role across Tech Lead, Architect, Security Gate, and Release) paired with strict sandbox separation: fail-closed JSON schema validation, exponential backoff, and hard isolation preventing judge roles from invoking shell tools.
- **Measured Impact:** 100% schema and policy compliance verified offline before any model enters a live 50/50 production trial.

### Zero-Trust Multi-Tenancy: Cryptographic State Seals & Privacy-Preserving Learning (Commit 2284da4)
- **Sample:** 14 Cryptographic Controls · Multi-Tenant Isolation Suite
- **What Failed:** First-generation agent frameworks commingle tenant state with orchestration code, store unencrypted snapshots on shared disks, and risk leaking proprietary domain literals or business logic into cross-project prompts during continuous learning.
- **Deterministic Fix:** Engineered end-to-end cryptographic isolation: per-workspace age-encrypted backups with automated 24-hour restore verification, HMAC-SHA256 state seals around every agent invocation, signed GDPR Art. 17 crypto-erasure certificates, and a deterministic literal refusal gate that blocks any rule containing tenant names, domains, or file paths from leaving the workspace.
- **Measured Impact:** Zero cross-tenant state leakage across continuous isolation probes; federated stack patterns verified across independent workspaces.

### Executive Command in Discord: Governing 32 Workflow Nodes in Under 15 Minutes a Week (Commit 11dc9dd)
- **Sample:** 11 Governance Primitives · Discord Command Relay
- **What Failed:** In conventional agent loops, humans are interrupted too late (after full implementation), chat feedback is lost between context windows, noisy notifications cause alert fatigue, and escalation prompts never quantify the cost of delay.
- **Deterministic Fix:** Built an audited operator-signal ledger injected live into every downstream model call, interactive Discord decision cards with 1-tap approvals, /hint and /rule slash commands, explicit wait-cost and default actions on every escalation, and a strict weekly interrupt budget.
- **Measured Impact:** Over 90% of operator responses directly alter downstream execution while requiring under 15 minutes per week of human review.

### 122 Best-Practice Rules at Zero Extra Tokens: Deterministic AST & Regex Auto-Healers for Dev, Design, and PM (Commit 0316ff9 → 5dd671a)
- **Sample:** 122 Rules · 3 Squad Roles · ~4,200 Tokens Saved per Feature
- **What Failed:** Injecting 100+ engineering, UX, and product rules into every agent system prompt wastes 3,000–5,000 input tokens per call and degrades instruction adherence. Worse, when a Product Manager or Designer emitted a weak acceptance criterion or sub-12px typography token, the pipeline triggered 2–3 extra LLM repair calls (~3,500 tokens each).
- **Deterministic Fix:** Engineered a 3-Layer Zero-Token Architecture (dev_rules.py, design_rules.py, pm_rules.py): Layer 0 runs deterministic Python auto-healers (<2ms, 0 LLM tokens) that rewrite weak Gherkin ACs, link (source: F#) research citations by lexical overlap, and upgrade sub-12px Tailwind classes in place; Layer 1 runs 0-token diff/AST/regex auditors; Layer 2 injects a <110-token file-extension or domain-routed micro-brief only when a retry occurs.
- **Measured Impact:** Net token-negative enforcement: ~65% token reduction in PM/UX node retries and 100% deterministic compliance across 122 rules.

### Anatomy of AI Citations: Why 72% of AI Recommendations Decay in 14 Days and How Dynamic Edge AST Serving Fixes It (Commit ab58d7a → 28ddb17)
- **Sample:** 5 Answer Engines · 12 Competitors · 240 Probed Prompts · 3-Hop Co-Citation Graph
- **What Failed:** Conventional SEO relying on static HTML sitemaps and bloated JSON-LD schema fails in agentic answer engines: 72% of LLM citations decay or get displaced within 14 days when competitors update their docs. Even worse, 34% of observed citations in raw LLM outputs attribute claims to stale URLs or entirely hallucinated source domains.
- **Deterministic Fix:** 1) Built a continuous multi-engine Share-of-Voice Radar (citation_engine.py / research_radar.py) measuring DOMINANT, CHALLENGER, and DISPLACED posture; 2) Deployed Zero-Cloaking Dynamic Edge Middleware serving LLM markdown mirrors matching production DOM with cryptographic integrity; 3) Automated 1-hop and 2-hop co-citation graph harvesting across high-authority ecosystem anchors (Reddit, GitHub, G2, Wikidata).
- **Measured Impact:** 3.4x higher citation persistence across Perplexity Sonar and SearchGPT, with automatic counter-positioning generated within minutes of citation loss.

### Continuous Negative Space Mining: Turning Competitor Pricing Canaries and Review Gaps into Shipped Code (Commit 28ddb17)
- **Sample:** 12 Competitor Registries · Automated Protocol Probes · 6 Thematic Clusters
- **What Failed:** Manual market research and quarterly analyst reports (Gartner, G2) lag market shifts by 90–180 days. Startups get blindsided when incumbents eliminate free tiers, double seat pricing, or launch competing protocol endpoints (FastMCP, A2A). Furthermore, user complaints on GitHub and Reddit remain unmined anecdotes rather than structured engineering tasks.
- **Deterministic Fix:** 1) Implemented deterministic HTTP Canaries (mira_intel.py / research_radar.py) that diff competitor starting prices, tier packaging, /llms.txt availability, and MCP endpoints on every research cycle; 2) Developed Lexical Negative Space Clustering that groups 1-star to 3-star reviews and competitor weaknesses into 6 core themes with verifiable citation anchors (source: F#); 3) Fed clusters directly into autonomous PRD and Roadmap synthesis nodes.
- **Measured Impact:** Zero-lag detection of competitor pricing hikes (e.g. Peec AI +101% price jump, free tier drop), and autonomous roadmap prioritization backed by 100% commit-verified customer pain evidence.


## 4. Portfolio Ventures
### OmniForce (https://www.omniforce.run)
- **Category:** Production Venture 01 · AI Search Visibility & GEO Platform
- **Isolation Unit:** projects/omniagent/ (Dedicated SQLite + age Vault + Isolated Worktrees)
- **Stack:** Next.js, TypeScript, FastMCP Gateway, Playwright E2E, Polar MoR
  - Operated 24/7 by Brevio Harness in continuous autonomous mode with 15-minute canary monitoring and visual regression checks.
  - 100-Point Agentic SEO engine, token-budgeted /llms.txt endpoints, and self-serve GEO audit pipeline shipped autonomously.
  - Primary production testbed for our 14-day, 58.2M-token engineering benchmark and 140+ Golden Evaluation contracts.

### EuroComply (https://eurocomply.app)
- **Category:** Production Venture 02 · European Regulatory Intelligence & MCP
- **Isolation Unit:** eurocomply/ (Canonical EUR-Lex Corpus + Streamable HTTP MCP v0.5.0)
- **Stack:** Next.js 16, React 19, Tailwind v4, Prisma + Postgres, MCP JSON-RPC 2.0
  - Serves 12 deterministic MCP tools at eurocomply.app/api/mcp with €0 inference cost and zero hallucinated statutory citations.
  - Powers real-time regulatory screening inside the Brevio Launchpad—classifying EU AI Act Annex III vs. Art. 50 obligations in <150ms.
  - Demonstrates cross-domain generality: the same autonomous harness operates both high-velocity MarTech and statutory legal infrastructure.


## 5. BYOK Pricing Tiers
- **Brevio Core (€0 — Free forever · Local CLI):** The deterministic single-task CLI state machine for individual engineers who want to inspect the architecture and run one slice at a time locally. [Token Policy: 100% Bring Your Own Keys (Zero markup)]
- **Brevio Harness Pro (€0 — until March 31, 2027 · then €199/mo):** Full Nova 0→1 Venture Architect + 24/7 autonomous PM, Design, Engineering, QA, and self-healing runtime for 1 product — free through Q1 2027 (March 31, 2027) in exchange for benchmark telemetry and product feedback. [Token Policy: 100% Bring Your Own Keys (€0 token markup)]
- **Brevio Studio (€0 — until March 31, 2027 · then €499/mo):** Complete autonomous Venture Studio: launch new 0→1 SaaS products with Nova and scale up to 5 repositories with the Autonomous AEO/GEO Growth CMO — free through Q1 2027 (March 31, 2027) for founding partners. [Token Policy: 100% Bring Your Own Keys (€0 token markup)]