a personal context engine

bavi

Always on. Captures what I see, hear, and say across every device. Recalls any moment on demand. Acts on what it learns.

capturing live
220K
ocr frames
114K
audio chunks
191K
embeddings
79
days running

recently from bavi

48 days · scroll for more

What Bavi noticed I worked on, summarized nightly and reviewed before publishing. The agent writes the bullets; I just approve them.

may 22
  • built bavi project page for yc startup school
  • helped business owner with their ai workflows
  • met with founder to list bavlio and bavimail on marketplace
may 21
  • worked on staging deployment for bavlio
  • conducted integration testing for front and back end
  • iterated on prompt design for improved performance
may 20
  • worked on optimizing bavimail integration
  • discussed improvements for bavlio features
  • refined job processing heuristics
may 17
  • addressed environment variable issues during merge
  • discussed codec functionality improvements
  • explored mentoring opportunities in tech
may 16
  • final preparations for bavlio launch on may 19
  • refined email testing process for baviMail
  • updated project metrics for better tracking
may 15
  • launched bavimail on product hunt
  • engaged with community feedback
  • monitored performance metrics
may 14
  • worked on bavimail email API enhancements
  • fixed glitchtip alert forwarding issues
  • planned content for second brain project
may 13
  • finalize bavimail launch preparations
  • review seo copy for bavimail
  • plan content for upcoming projects
may 11
  • updated bavlio api docs for agent-first architecture
  • created investor pitch deck for bavlio and bavimail
  • pushed pr 71 to bavimail and confirmed
may 10
  • rest day
may 9
  • update bavlio api docs for agent-first architecture
  • debug null reference error in api-kevs.tsx
  • compress pdf files for oci application
may 7
  • updated bavlio api docs
  • worked on capcut media features
  • tested google chrome functionalities
may 6
  • debugged personalization playground issues
  • reviewed pr for iTerm2 enhancements
  • worked on CapCut video edits
may 5
  • worked on capcut video editing features
  • optimized bavi for better performance
  • merged pr for find leads page alignment
may 4
  • optimized bavlio mvp campaign worker
  • improved log hygiene and structlog
  • worked on capcut video features
may 1
  • worked on BaviMail feature improvements
  • refined openClaw automation capabilities
  • iterated on SapienEx user experience
apr 30
  • refined api key management process
  • optimized email finder functionality
  • enhanced front-end testing capabilities
apr 29
  • worked on oauth flow improvements
  • connected gmail and google drive
  • discussed video content structure
apr 28
  • debugged capture pipeline outage
  • updated campaign-playground remediation
  • worked on BaviMail indexing issues
apr 27
  • worked on BaviMail integration improvements
  • discussed features for SapienEx updates
  • refined user prompts for openClaw
apr 25
  • rest day
apr 24
  • worked on bavi email warm-up features
  • explored list management integration
  • reviewed openai's latest model updates
apr 20
  • debugged jwt validation for staging
  • worked on bavlio admin features
  • brainstormed resume and portfolio updates
apr 19
  • worked on exam preparation course structure
  • refined content for reddit drafts
  • optimized bavlio outreach strategy
apr 18
  • worked on stripe promotion codes integration
  • validated csv upload functionality
  • prepared for upcoming production deployment
apr 17
  • worked on csv upload validation hotfix
  • polished lead finder ux
  • updated bavi prompt for better performance
apr 16
  • deployed updates to production
  • worked on integration for Bavlio Mail
  • conducted smoke tests on production environment
apr 14
  • debugged omi app error on mobile
  • worked on claude demo video script
  • optimized bavlio admin interface
apr 13
  • worked on bavlio mail features
  • collaborated on video scripts
  • discussed infrastructure setup
apr 11
  • tested pixel deployment on bavlio
  • reviewed documentation for sapienex
  • collaborated on project planning with team
apr 10
  • worked on bavi context engineering infrastructure
  • updated bavi repository with recent changes
  • discussed data ingestion formats for ai agents
apr 9
  • worked on sapienex project initiation
  • optimized memory usage in bavlio admin
  • answered questions in research project
apr 8
  • conducted seo audit for bavlio python agent
  • worked on bavimail onboarding process
  • optimized analytics dashboard for better insights
apr 7
  • completed project 4 file system assignment
  • continued ai workforce architecture research
  • worked on bavlio teaser draft
apr 6
  • pushed commit for project 4 file system
  • worked on marketing strategy for BaviMail
  • analyzed audio insights for ongoing project
apr 5
  • worked on bavimail deliverability features
  • completed project 4 file system assignment
  • discussed marketing strategy for bavimail
apr 4
  • completed project 4 file system assignment
  • filled design document template
  • added subdirectory support
apr 1
  • developed new features for bavi integration
  • optimized expense tracking in sapienex
  • refined outreach workflow in bavlio
mar 31
  • worked on expense dashboard for sapienex
  • refined admin features in sapienex
  • continued fundraising efforts for bavi
mar 30
  • installed python-docx and reportlab
  • developed karaoke system song library
  • enhanced Bavlio Mail accessibility
mar 29
  • installed new libraries for bavlio-admin
  • worked on expense tracking features
  • updated project documentation
mar 28
  • reviewed bavlio feature updates
  • optimized bavi performance metrics
  • addressed user feedback on sapienex
mar 24
  • optimized bavi performance metrics
  • merged new features into bavlio
  • discussed api improvements for sapienex
mar 23
  • worked on bavlio mail accessibility
  • refined sapienex consulting processes
  • developed deep research prompt v2
mar 21
  • updated sapienex prd for human readability
  • integrated real-time sports results into sapienex
  • verified bavlio-prod container health
mar 20
  • worked on bavimail email platform enhancements
  • continued development on sapienex integration
  • debugged issues in openclaw deployment
mar 18
  • added unit tests for url unwrapping
  • worked on custom smtp integration
  • refined multi-device architecture docs
mar 17
  • fixed issues in bavlio python agent
  • enhanced deliverability dashboard in bavimail
  • worked on sector growth features for sapienex

what it sees

Screen

3 monitors · 1.2 fps

Every window, every app, every URL. OCR via Screenpipe; metadata via macOS Accessibility.

Audio

system + mic · diarized

Meetings, calls, music. Whisper transcribes; pyannote separates speakers.

Pendant

Omi · BLE always-on

Conversations away from the desk. Streams to phone, syncs to the same DB.

architecture

Capture happens locally. Text, transcripts, and metadata stream to Azure. Raw images and audio purge after sync so the laptop stays clean.

Screen3 monitorsAudiosystem + micPendantOmi BLEsync daemonevery 10 minutesazure pgvectorvoyage-3-large embeddingsclaude code /bavisemantic recalllocal media purges after sync
screen capture
audio capture
pendant capture
sync + storage

ask bavi

Every question gets answered by a Claude Code skill that pulls from the embeddings, the transcripts, and the OCR text in one query.

claude code · /bavi
>
searching ocr_frames · audio_chunks · summaries
cosine match: 14 chunks · 3 days · 2 speakers
ranking by recency + semantic relevance

it also acts

Recall is half the value. Bavi runs a small fleet of agents on its own context, on its own schedule. They write to Telegram, Gmail, Google Calendar, and the laptop directly.

Morning briefing

06:00 ET · Telegram

Bavi reads overnight conversations, calendar for the day, and any unanswered messages, then sends a one-screen brief.

Inbox triage

02:00 ET · Gmail

Classifies overnight email with a GPT-5-mini agent, drafts replies for the easy ones, surfaces what needs a human.

Calendar logging

nightly · Google Calendar

Extracts every commitment I made out loud during the day and back-fills events so my calendar matches what actually happened.

Proactive nudges

sentinel loop

Watches for patterns: 'you mentioned following up with X three days ago.' 'You said you would ship Y by today.' Pings Telegram.

Voice chat

instant · WebRTC

Pipecat realtime voice agent. I talk to Bavi out loud on a walk, it answers with the same context the Claude Code skill uses.

Self healing

always

If Screenpipe stalls, mic dies, or the pendant disconnects, the supervisor agent diagnoses and restarts without me noticing.

capture layers

Coverage grows over time. Every new layer feeds the same database, so the agent gets richer without re-architecting.

Desktop

live

MacBook Pro · 3 monitors · system + mic audio

Screenpipe captures OCR text at 1.2fps across all displays. Mic and system audio transcribed via Whisper. Synced every 10m.

Wearable

live

Omi pendant · BLE → phone → cloud

Always-on mic streams audio from anywhere. Multi-speaker diarization separates Adam from others.

Mobile

next

iOS screen + ambient audio

Phone screen capture and ambient mic for context away from the desk. Custom iOS app in design.

Integrations

future

Calendar · Email · Slack · iMessage · Git

Direct API ingestion so structured data does not have to round-trip through screen OCR.

stack

Capture

  • Screenpipe
  • Omi
  • macOS HealthKit

Transport

  • launchd
  • Python sync daemon
  • Cloudflared tunnel

Storage

  • Azure Postgres
  • pgvector
  • Azure Blob

Intelligence

  • Voyage-3-large
  • Whisper
  • pyannote diarization

Query

  • Azure Container Apps
  • Claude Code skill
  • Telegram bot