Aequitas · Latin · Justice, Equity, Precision

It remembers your intent.
So you never solve
the same problem twice.

Aqueitas intercepts your engineering work at the moment it happens, reasons over it, and makes every decision permanently queryable. Your knowledge compounds. Your mistakes don't repeat.

View on GitHub See the Architecture
aqueitas · sensor active
→ git commit -m "fix: resolve OOM error in CF Node.js module via memory cap"
──────────────────────────────────────────────
[Aqueitas] Intercepting diff...
[Aqueitas] Routing to Hybrid Context Engine...
[Aqueitas] Reasoning over intent via DeepSeek...
✓ Intent: Memory cap applied to prevent OOM in Cloud Foundry
✓ Vectorized → Written to Sovereign Vault [1536d]
──────────────────────────────────────────────
[master 4f3a2c1] fix: resolve OOM error in CF Node.js module
→ aq-ask "how did we fix the Cloud Foundry memory issue"
──────────────────────────────────────────────
Resolution found in Vault [similarity: 0.97]
The Node.js CF module was hitting OOM errors under load.
Resolution: applied memory: 512M cap in mta.yaml,
preventing heap overflow at runtime.
Committed: 3 weeks ago · Project: BTP-Prod
System Architecture

Four phases. Zero friction.

Aqueitas operates through a strict event-driven intercept architecture. Every layer has a single responsibility. Nothing is redundant. The entire stack runs locally — no SaaS, no lock-in.

I
The Sovereign Vault
PostgreSQL 17 with pgvector. HNSW indexing on 1536-dimensional embeddings. Your data, locally owned, permanently indexed.
II
The Ingestion Engine
FastAPI service with hybrid AI routing. DeepSeek extracts intent. OpenAI encodes it mathematically. Two models, two jobs, no overlap.
III
The Sensor
A global Git hook at core.hooksPath level. Intercepts every commit synchronously. You never leave the terminal.
IV
The Retrieval Engine
RAG pipeline via aq-ask CLI. Vector proximity search with a Zero-Hallucination Protocol — if it's not in the Vault, it says so.
The Origin

Built from the rubble
of a grander failure.

Aqueitas did not begin as Aqueitas. It began as an obsession — and an instructive collapse.

Act I · 2024
Project Jarvis
A personal AI orchestration system. Self-hosted n8n on AWS polling LinkedIn, Gmail, Teams — routing everything through agents to build a live decision dashboard. The vision was total. The complexity collapsed it.
Instructive Failure
Act II · 2024
Aqueitas 1.0
The lesson was precision over scope. Strip the problem to its core. Git diff interception, Gemini CLI intelligence, Notion persistence via MCP. Narrow, focused, shipped.
Shipped
Act III · 2025
Aqueitas 2.0
Architectural maturity. Sovereign infrastructure, hybrid AI context engine, HNSW vector search, zero-hallucination RAG. Production grade at under $1/month operational cost.
Live
$1
Total monthly API overhead for 400 dense intelligence operations.
DeepSeek reasoning at $0.14/1M tokens. OpenAI embeddings at $0.02/1M tokens.
Scale-to-zero architecture. Zero idle compute cost.
Philosophy

Efficiency over Excess.
Sovereignty over Convenience.

Zero-Cost at Idle
The system avoids managed databases and always-on cloud compute. Local containerization for development, with a clear path to AWS Lambda (scale-to-zero) for production. You pay for nothing when you're not working.
Zero Hallucination
The retrieval layer is explicitly forbidden from utilizing generalized training data. If the Vault contains no record of a resolution, it responds exactly: "The Vault contains no record of this resolution." No fabrication.
Data Sovereignty
Your intellectual property belongs to you, not a SaaS vendor. Every thought, every decision, every architectural rationale lives in your own PostgreSQL instance. Portable. Private. Permanent.
The Death of Passive Forgetting
Engineers re-solve the same problems across teams, across careers. Aqueitas makes every hard-won resolution queryable forever. The compounding effect of retained intelligence is the actual competitive advantage.
Built by

Siddharth Shrivastava

SAP BTP & Cloud Architect. AWS Solutions Architect. Mumbai.

GitHub Repository LinkedIn