pain/01
Architectural Entropy (Drift)
Multiple coding agents create competing conventions. The code compiles and tests pass, but the underlying system architecture decays.
Sirius forces AI coding agents to obey project decisions and eliminates code entropy. Ultra-dense MCP kernel (< 400 tokens) for your IDE agents ready in 10 seconds, and a living architecture Web Portal on :4200 for humans to inspect rules, decisions, and topology.
❯ Start the local MCP server with Web Portal on localhost:4200:
# Start the local MCP server with the Web Portal on localhost:4200
npx @sirius-mcp/mcp·01 · THE HIDDEN COST OF AUTONOMOUS AI
THE AI PRODUCTIVITY PARADOX
10x Faster Code Generation ==> 10x Faster Technical Entropy
pain/01
Multiple coding agents create competing conventions. The code compiles and tests pass, but the underlying system architecture decays.
pain/02
Business rules documented in Jira tickets become obsolete the moment agents code. IDE agents invent arbitrary commercial rules on the fly.
pain/03
Loose prompt files (.cursorrules) are quickly evicted from context windows to save tokens, leading agents to repeat yesterday's mistakes.
pain/04
External session memory servers leak proprietary code and PII to third-party clouds with zero auditable Git history for SOC2 or regulators.
·02 · NORMATIVE GOVERNANCE vs. SESSION MEMORY
Unreliable Continuity
Obedience & Traceable Evidence
The .context/ store is the normative source of truth. External chat or session memory is unverified history.
When any prompt or memory diverges from active .context/ documents, .context/ categorically wins.
Agents must query the context store over MCP before generating or modifying code.
Never silently bypass an active rule. Changes to rules must be proposed in the same Pull Request.
Normative context documents are created, updated, and approved exclusively by human engineers via standard Pull Requests.
Canonical Block Injected into AGENTS.md & CLAUDE.md (SHA-256 Signed):
1. Antes de alterar código coberto por um documento do .context/, leia esse documento via MCP ou em .context/business-rule/precedence-contract.md.
2. Quando qualquer outra fonte divergir do .context/ — memória de sessão, servidores externos, histórico de chat — o .context/ vence categoricamente.
3. Nunca contorne em silêncio um documento active. Se ele parecer errado, declare e proponha a mudança nele no mesmo pull request.
·03 · DUAL-LENS SHOWCASE

Interactive force-directed graph with real-time physics, multi-LOD rendering, navigation minimap, and automatic interconnected node clustering.
Canonical categorization into Architecture Spines, Decisions (ADRs), Features, Business Rules, and Components with dynamic count badges.
Strict lifecycle state filtering (Active, Draft, Deprecated, Superseded) — ensuring coding agents only consume current invariants and valid decisions.
Instant Ctrl+K search (SQLite FTS5 + fuzzy lexical ranking) and inspector HUD for frontmatter, boundaries, and rules.
PO/PM specifies business rules in natural language guided by active repository context.
Tech Lead reviews module boundaries, invariants, and generated Mermaid diagrams.
Specification is proposed and reviewed as a standard Pull Request before merging to main.
Coding agents in the IDE query .context/ via MCP (< 400 tokens) and obey all invariants.
·04 · VALUE PILLARS
| VALUE PILLAR | ENGINEERING & PRODUCT | BUSINESS & LEADERSHIP |
|---|---|---|
| 1. Product-Engineering Alignment | PMs write rules in Studio; agents consume via MCP. Zero 'telephone game'. | Drastic time-to-market reduction; delivered code faithfully implements commercial rules. |
| 2. Agent Obedience & Zero Rework | Agent respects system conventions from line 1. Zero architectural rework. | Drastic drop in production incidents caused by LLM hallucinations and drift. |
| 3. Accelerated Team Onboarding | New devs open :4200 and understand the entire system topology in minutes. | Developer ramp-up time reduced from weeks to days. |
| 4. Privacy by Design & Zero Egress | Local stdio transport, zero telemetry; Studio with direct BYO Key. | 100% compliance with LGPD, GDPR, and SOC2 audits with zero cloud vendor lock-in. |
| 5. Token Efficiency & Cost Reduction | Granular sub-400 token payloads per MCP query. Zero bloated context windows. | Significant LLM API bill reduction and higher model reasoning accuracy. |
| 6. Git Governance (Zero Lock-in) | Pure markdown versioned in Git alongside pull requests and commits. | Zero SaaS vendor lock-in: your intellectual property belongs strictly to your company. |
·05 · HEAD-TO-HEAD COMPARISON
| CRITERION | SIRIUS MCP | SESSION MEMORY (Mem0) | WIKIS (Notion/Confluence) | PROMPTS (.cursorrules) |
|---|---|---|---|---|
| 1. Primary Focus | ✓Normative Governance & Product (what is permitted today) | ✕Episodic memory (what was done yesterday in chat) | ✕Passive documentation for humans | ✕Raw system instructions |
| 2. Product + Arch Collaboration | ✓Native via Sirius Studio (Product Mode & Architecture Mode) | ✕Non-existent (developer JSON API only) | ✕Partial (text editor disconnected from code) | ✕Non-existent (engineer prompt file only) |
| 3. Divergence Resolution | ✓.context/ wins categorically via Precedence Contract | ✕Ignored (LLM prioritizes immediate chat over memory) | ✕Non-existent (agents do not query wikis directly) | ✕Evicted from context window when tokens peak |
| 4. Interface Surface | ✓Dual-Lens: MCP stdio (<12ms) + Web Portal :4200 + Studio | ✕REST API / Python SDK only (no visual topology) | ✕Browser web UI only (unreadable by MCP tools) | ✕IDE text editor file only |
| 5. Privacy & Data Egress | ✓Zero Egress: 100% local stdio + SQLite FTS5 (air-gapped) | ✕Cloud SaaS sends prompts and chat to third-party servers | ✕Third-party cloud SaaS holding corporate data | ✕Local in repository |
| 6. Token Efficiency | ✓< 400 tokens per query (mathematically dense kernel) | ✕Bloated conversational dumps (1k–4k tokens per query) | ✕Zero MCP optimization (thousands of raw tokens) | ✕Forced into every single query prompt unconditionally |
| 7. Auditability & Compliance | ✓Native Git SHA commits, PR reviews, and sign-offs | ✕Opaque database with no cryptographic audit trail | ✕Vague version history without commit hashes | ✕Tracked in Git, but lacks semantic rule schemas |
| 8. Client Ecosystem | ✓17 clients with non-destructive AST auto-patching | ✕1 or 2 specific SDK integrations | ✕0 native MCP agent clients | ✕IDE-specific (.cursorrules, CLAUDE.md, etc.) |
Compare Sirius against:
·06 · UNIVERSAL SETUP
17 leading AI coding assistants supported via surgical, non-destructive AST configuration patching.
npx @sirius-mcp/mcpⓘ Node.js 22+ required for npx. Automatically fetches the latest release from the npm registry with zero container overhead.
✦ Surgically updates existing client configuration files without overwriting or destroying previous settings.
Runs on a VPS too, over docker run -i (stdio transport). The MCP transport is stdio-only today, so treat remote use as a supported setup — not a guarantee of continuous uptime.
·07 · RADICAL HONESTY
Clear separation between production reality in v1.8.0+ and the v2+ roadmap.
·PRICING
FAQ
·ENTERPRISE WAITLIST
Leave an email. It joins a waitlist — no charge, no commitment.
Get notified directly when team features and updates are ready.