The research company behind QIA and PayQIA.
Haqikos exists because every AI agent — no matter how capable in a single session — forgets everything when that session ends. Developers rebuild memory logic from scratch, again and again, with no shared standard and no persistent layer beneath the LLM.
Work on this problem started in 2024. The first thing we built was QIA — a context and intelligence engine designed to handle semantic search, context compression, and automatic memory decay properly, rather than offloading those decisions to the calling developer every time. QIA became the foundation. PayQIA is the first product shipped on top of it.
The three-layer structure — Haqikos (company) → QIA (engine) → PayQIA (product) — exists for a specific reason: keeping the research layer separate from the shipped product means QIA can evolve as a proper engine family, with future products built on it, rather than being permanently coupled to one product's constraints. The same reason Anthropic keeps Claude the model separate from Claude.ai the product.
The team
Haqikos is a small, focused team. We are based in New Delhi, India, operating as a primarily remote-first engineering and research group. We don't list team members by name on the public site by default — if you're a developer, investor, or researcher looking to work with us directly, contact us and we'll respond personally.
Founded
2024
New Delhi, India
Team size
Small
Focused, not padded
Primary focus
QIA + PayQIA
One engine, one product
Timeline
Founded
Haqikos established in New Delhi, India. Work begins on the core problem: AI agents that lose all context when a session ends. Early research into persistent vector memory and retrieval efficiency.
QIA Core Vector DB
First internal version of QIA's vector memory layer. Benchmarking begins on context compression — how much of a session history can be safely discarded without losing signal.
PayQIA Alpha
First developer-facing product built on QIA. SDK shipped in TypeScript, Python, and cURL. Internal testing with real agent workloads begins.
PayQIA Live
PayQIA launches publicly as the Memory OS for AI agents. Free, Pro, Scale, and Enterprise tiers available. NEWQIA Vector Memory becomes the default retrieval backend.
Engineering Expansion
Core engineering team expands in New Delhi to accelerate QIA's memory infrastructure and the PayQIA developer platform.
Now
QIA v1.0 in production. Active research on context compression, knowledge graph entity resolution, multi-agent shared memory, and memory-system safety. Compliance roadmap (SOC 2 Type II) in progress.
How we work
We label unverified claims as unverified.
Every benchmark number on haqikos.ai and payqia.com is marked as internal, pre-launch, or independently verified — nothing in between. We carry this from PayQIA's own disclosure standard.
Research and product inform each other directly.
QIA's retrieval architecture is shaped by what we're learning in research on context compression and graph decay — not by what sounds good in a press release. The Systems Lab and product team are the same people.
Memory infrastructure must be deletable, not just fast.
Speed without control is a liability in production. Every memory stored in QIA can be explicitly deleted by the developer at any time. The right to forget is an API primitive, not a support request.
Location
New Delhi, India
Engineering base. Remote-first for most roles. For enterprise, partnership, or press inquiries, we respond to direct emails.