Personal AI Memory System

Your context.
Your file.
Your control.

OpenYantra is a local-first memory standard for one human and their personal AI agents. One human, one file, full auditability, and a calm briefing-room interface over a human-readable ledger.

1
Memory file
14
Domain sheets
100%
Local-first
MIT
License
~/Chitrapat.ods Active
Memory
Ledger
Loops
Agents
Episodic
247
Semantic
133
Prospective
89
Beliefs
41
12
Open Loops
4
Agents
3
Inbox
OK
Health
02 / Architecture

Built on non-negotiable invariants.

OpenYantra enforces six architectural invariants that can never be broken. Every feature must pass all six. Complexity that violates any invariant is rejected at the design stage.

01
Single Memory File
All persistent memory lives in Chitrapat.ods. Human-readable, portable, directly editable. Never replaced by databases or binary formats.
Canonical Source
02
Single Writer Rule
Only Chitragupta writes to memory. All other agents are read-only. Every write flows through the Sanchitta queue — no exceptions.
LedgerAgent
03
Domain Sheet Model
Memory is structured across 14 domain sheets: Identity, Goals, Projects, People, Preferences, Beliefs, Tasks, Open Loops, Session Log, Agent Config, Ledger, Inbox, Security Log, and Quarantine.
14 Domains
04
Full Auditability
Every write is logged, timestamped, and SHA-256 signed. The Ledger is append-only. Audit capability can never be removed or bypassed.
SHA-256
05
Local-First
Runs fully offline. No cloud dependencies, no external APIs required. External LLMs are optional, never required. Your data never leaves your machine.
Offline
06
Lightweight Retrieval
VidyaKosha uses BM25 and TF-IDF tuned for 100–5000 records. No vector databases, no heavy infrastructure, no magic.
VidyaKosha
Write flow — the only path to memory
01 / Source
Agent
Any AI agent
›
02 / Queue
Sanchitta
WriteQueue
›
03 / Writer
Chitragupta
LedgerAgent
›
04 / Store
Chitrapat.ods
Canonical memory
03 / CLI Interface

One command.
Zero friction.

The CLI is the primary interface. Every command is short, predictable, and scriptable. Capture ideas, inspect memory, route queries, and monitor health from your terminal.

yantra inbox "idea"Capture to queue›
yantra loopsShow open loops›
yantra healthFile + ledger check›
yantra diff --since yesterdayMemory delta›
yantra route "query"Route to best agent›
yantra recall "topic"BM25 retrieval›
yantra — zsh
$yantra inbox "async write batching idea"
Queued WriteID: #a3f9d2c1 → Sanchitta
$yantra loops
12 open loops found
Oldest: "refactor VidyaKosha BM25" — 14 days
$yantra health
Chitrapat.ods OK — 1.2MB
Ledger: 489 entries — Sigs: valid
Last write: 4m ago by Chitragupta
$yantra diff --since yesterday
+7 records written
2 tasks closed, 1 belief updated
$
04 / Agent Ecosystem

Many agents.
One memory.

Multiple AI agents share a single memory file. Each has a defined role: read, write, or route. The architecture prevents conflicts, enforces auditability, and keeps the human in control of every record.

Agents — Read Only
Retrieval
VidyaKosha
BM25/TF-IDF memory search and context retrieval
Read
Routing
Sutra
Routes queries to the best available agent
Read
Planning
Karma
Task and open-loop management agent
Read
Capture
Inbox Agent
Receives and queues raw inputs from the user
Queue
↓ All writes routed through
Sole Writer
LedgerAgent
Chitragupta
The only agent with write access. Signs every record with SHA-256. Enforces the single-writer rule across the entire system.
Write
↓
Chitrapat.ods
Canonical Memory — Local File
Source of Truth
Read Only
Write (Chitragupta only)
Queue / Inbox
Verified / Signed
01
Simple
One file. One writer. No hidden side effects. Complexity is rejected, not deferred.
02
Inspectable
Every record is human-readable. Every write is auditable. You own the full stack.
03
Local
Runs fully offline. No telemetry. No cloud. Your cognition, on your machine.
04
Human-Controlled
The .ods file is always editable by you. AI assists. You decide what persists.
05 / Live Simulation

See it
in action.

Simulate the yantra CLI without installing anything. Click a command or type your own to see how OpenYantra responds. The memory state updates with every write.

yantra — simulation
Records247
Open Loops12
Ledger Entries489
HealthOK
Last Write4m ago
06 / Release Notes

Current release.
Shipped as one audited bundle.

OpenYantra versions ship as complete, user-visible releases. The landing page, dashboard, runtime, and documentation move together so the published state always matches the product you download.

v2.11 Released: March 17, 2026

Briefing Room Dashboard Overhaul

  • Rebuilt the dashboard in the new v3 Briefing Room design language.
  • Introduced a 7-tab sidebar: Today, Inbox, Loops, Projects, Timeline, Review, and System.
  • Added a collapsible sidebar, light/dark mode toggle, and full mobile responsiveness down to 375px.
  • Migrated the v2.10 dashboard behaviors into the new UI, including Today, Inbox, Loops, Projects, Timeline, Review, System, and Copy Context.
  • Added the Daily Insight panel with a stable oracle-card hook for Oracle proactive surfacing.
  • Switched yantra_ui.py to serve UI/v3/dashboard.html via FileResponse.
  • Bumped the OpenYantra core and dashboard version to 2.11.
  • Updated the landing page to v2.11 and added a Docs dropdown linking the Brand Manual and Visual Guide.
  • Refreshed the README and deployment docs for the v3 UI architecture.