VERITY

Open-source agent for knowledge-dense lives

Made for knowledge workers · research & discovery · Telegram + Obsidian

Research, collect, recall, track habits, process voice — all from Telegram, integrated with Obsidian. Self-hosted, open-source.

View on GitHub

/claude Compare REST vs GraphQL for a mobile app with offline support 14:23

👍1

WebSearch "REST vs GraphQL offline mobile 2026" Read graphql.org/comparison Read sqlite.org/sync

REST vs GraphQL for offline mobile:

REST: simpler caching with HTTP, easier offline sync via ETags
GraphQL: flexible queries, but caching is harder
For offline-first: REST + SQLite is the pragmatic choice

14:23

What about conflict resolution? 14:24

👍1

Read crdt.tech/implementations Read yjs.dev/docs

For conflict resolution with offline sync:

Last-write-wins — simplest, data loss risk
CRDTs — automatic merge, complex to implement

I'd recommend CRDTs with Yjs if real-time collaboration matters.

2m 8s · 3 reads · 4 commands
Session: ●● a3f7c1e2

🔄 Retry ▶️ More 🆕 New

14:24

/research Current state of local-first software in 2026 10:15

👍1

WebSearch "local-first software 2026" WebFetch automerge.org/blog WebFetch electric-sql.com/docs WebFetch zerosync.dev Write local-first-2026.md

Local-First Software in 2026

Key developments:

CRDTs have matured — Automerge 3.0 and Yjs dominate
SQLite-based sync (cr-sqlite, PowerSync) replacing traditional backends
Zero framework by Rocicorp gaining enterprise adoption
Apple adding CRDT primitives to CloudKit

Saved to your vault as local-first-2026.md 10:17

PDF version:
local-first-software-2026.pdf

4m 31s · 4 reads · 6 commands
Session: ●● 7b2e9f4d

🔄 Retry ▶️ More 🆕 New

10:17

/collect https://arxiv.org/abs/2401.15947 16:42

👍1

WebFetch arxiv.org/abs/2401.15947 Read vault/agents/*.md (12 matches) Write agent-effectiveness-paper.md

Saved to vault:

"The Unreasonable Effectiveness of LLM Agents" (2024)

Key claim: tool-augmented agents outperform fine-tuned specialists
Tagged: #agents #research #llm
Linked to 12 existing notes

Created agent-effectiveness-paper.md with summary, key quotes, and backlinks. 16:43

Also save this thought: agent reliability scales with tool quality, not model size 16:44

👍1

Added to today's daily note with link to the paper.

1m 47s · 2 reads · 3 commands
Session: ●● e5c8a1b3

16:44

/note flash-attention-v3 09:30

👍1

flash-attention-v3.md

FlashAttention-3 benchmarks from the paper. IO-aware attention with 2x speedup on H100...

Tags: #attention #performance #transformers
Backlinks: [[attention-is-all-you-need]], [[transformer-optimization]] 09:30

/note daily 09:31

👍1

20260223.md

Morning review: finished FlashAttention benchmarks. Meeting with design team at 14:00. Need to review PR #47...

3 tasks pending · 2 habits logged 09:31

Morning check-in

Meditation: done? · Exercise: done? · Journaling: done?
Meds: Vitamin D, Omega-3

Current streak: 12 days 08:00

✅1

Read vault/projects/local-first-app.md WebSearch "CRDT conflict resolution 2026" WebFetch electric-sql.com/blog/latest

New research on your topic

Found 3 new developments related to your local-first-app project:

ElectricSQL released v0.9 with full Postgres sync
New CRDT benchmarks show 10x improvement in merge speed
Expo SDK 52 adds built-in SQLite sync

Saved summary to local-first-updates.md 08:15

🧠1🔥1

Claude New Collect Settings Menu Sessions

Write a message

Frequently Asked Questions

Is Verity a wrapper around ChatGPT?

No. Verity is built on Anthropic's Claude Agent SDK with Claude Opus. It's not a simple API wrapper — it uses multi-step reasoning, tool use, and maintains persistent context. It can execute multi-turn research tasks, read and write to your knowledge base, and chain multiple tools autonomously.

What knowledge base does it connect to?

Verity integrates with Obsidian vaults via local sync or git. It can read your notes, create new ones, manage daily journals, and build backlinks. The semantic search uses vector embeddings so you can query by meaning, not just keywords.

Is my data private?

Your vault data stays on your infrastructure. Verity accesses it through secure APIs — nothing is stored on third-party servers. Conversations with Claude follow Anthropic's data handling policies (no training on your data with the API).

How do I deploy it?

Three commands: git clone, cd, ./install.sh. The script detects Python 3.10+, creates a virtualenv, installs dependencies, and runs the interactive setup wizard (bot token, API keys, optional features). Re-run anytime with --setup to reconfigure. Also supports Docker, systemd, and launchd deployment.

What does it cost to run?

The main cost is Claude API usage. Typical daily usage (10-20 interactions) costs around $1-3/day with Claude Sonnet, more with Opus. Web search and voice transcription have minimal additional costs. Server hosting is ~$5/month.

How is this different from Telegram's built-in AI?

Telegram's built-in AI is generic and stateless. Verity is personal — it knows your projects, your notes, your preferences. It can research the web, write to your vault, process voice messages into structured notes, and chain multiple actions. It's an agent, not a chatbot.

VERITY

What is Verity?

Three lines to your own agent

Claude Code Sessions

75+ LLM Providers

Obsidian Integration

Deep Research Pipeline

Voice & Media

Accountability System

Privacy & Data Control

Self-Modifying

Production Ready

/transform