Solo-founder alpha · open source · MIT
Your evolving
cognitive partner.
An open-source cognitive-assistant alpha built around memory, reasoning, and optional grounding — with its current limits stated plainly.
Most assistants forget who you are, hallucinate when they're unsure, and stagnate the day they ship.
Tanya is being built to remember, ground, and improve through reviewed releases.
Six systems that make Tanya
a cognitive partner, not a chatbot.
Each one is a workspace package you can read, fork, and self-host. Nothing is magic.
- 01 / Memory@tanya/ambient-memory
Memory architecture for continuity
The codebase contains working, episodic, semantic, procedural, social, biographical, and meta-memory components. Durable cross-session and cross-surface continuity remains an alpha property to verify per deployed surface — not a blanket guarantee.
working · episodic · semantic · procedural · social · biographical · meta - 02 / Safety@tanya/anti-hallucination
Optional source-grounding
When the grounding path runs, it extracts claims, checks available evidence, and can attach sources and confidence metadata. It is not guaranteed to run on every turn, and missing evidence is never presented as proof.
- 03 / Omnipresence@tanya/channels
Surfaces under controlled rollout
A standalone iMessage integration is implemented in the repository. Its channel effects remain disabled by default during WAVE-F0, like every bridge. Additional bridges and shared identity-memory continuity are development targets, not public availability claims.
- 04 / Truth@tanya/consistency-checker
Evidence when available
Grounded paths can expose sources and confidence metadata. An ordinary model answer is not a receipt, and Tanya does not claim that every response is independently verified.
- 05 / Reasoning@tanya/agi-runtime
Multi-agent reasoning components
Specialised planning, critique, synthesis, and verification components can collaborate under bounded runtime paths. User-facing evidence summaries may be emitted, but raw private reasoning and a trace for every answer are not promised.
- 06 / Sovereignty@tanya/security-blueprint
Self-host the open-source core
Inspect and run the MIT-licensed source on infrastructure you control. A managed cloud service is planned but is not commercially open during WAVE-F0.
Three steps. That's it.
Inspect or run the alpha
Explore the source and configure only the components and providers you explicitly choose, or join the waitlist for a future managed alpha.
Use scoped components
Memory, multi-agent reasoning, and source-grounding paths are available as alpha components, with behavior determined by the deployed configuration.
Inspect emitted evidence
When a grounded path emits sources or confidence metadata, inspect them. Answers without that evidence remain unverified model output, not proof.
“An AI that forgets who you are every morning is not an assistant. It is a stranger wearing a familiar voice.”
Read the full manifesto →
The alpha is small.
Make it smaller with you in it.
Tanya is in solo-founder alpha, with its launch limits stated honestly. Leave your email to request future updates; confirmation email remains disabled during WAVE-F0.
No spam. No growth hacks. One sprint, one email.