LettaBot

Overview

Status (June 2026): Archived. The GitHub repository was archived in May 2026 with the note "has been replaced by Letta Code channels/schedules."^[1] Existing deployments continue to work, but Letta directs new projects to the Channels feature built into Letta Code.^[2]

LettaBot was a personal AI assistant built on the Letta platform (formerly MemGPT), focused on persistent memory across multiple channels.^[1] Unlike most AI assistants that forget between sessions, LettaBot maintained a single unified memory across Telegram, Slack, Discord, WhatsApp, and Signal — remembering conversations from days, weeks, or months ago.

Built by the Letta AI team, it represented the "memory-first" approach to personal AI assistants. Its short life — first alpha in February 2026, final release in March, archived by May — ended not in abandonment but in absorption: the multi-channel idea was folded directly into Letta Code as Channels and Schedules.

Key Stats

Architecture

Telegram ──┐
Slack ─────┤
Discord ───┼──→ ONE AGENT ──→ ONE CONVERSATION
WhatsApp ──┤     (memory)      (chat history)
Signal ────┘

Single agent, unified memory. Start on Telegram, continue on Slack, pick up on WhatsApp — the agent remembers everything.

Features

• Multi-Channel — Chat across 5 platforms with unified context
• Persistent Memory — Remembers across days/weeks/months, not just sessions
• Voice Messages — OpenAI Whisper transcription across all channels
• Heartbeat — Periodic check-ins where agent reviews tasks
• Scheduling — One-off reminders and recurring tasks
• Local Tool Execution — Read files, search code, run commands
• Streaming Responses — Real-time updates as agent thinks
• Skills — ClawHub and skills.sh compatible

Installation

git clone https://github.com/letta-ai/lettabot.git
cd lettabot
npm install && npm run build && npm link
lettabot onboard
lettabot server

Requires Node.js 20+ and a Letta API key from app.letta.com. The repository is archived but remains cloneable; existing deployments continue to function.

The Letta Platform

LettaBot is built on Letta, the platform for stateful agents.^[3] Letta (formerly MemGPT) pioneered the concept of AI with advanced memory that can learn and self-improve over time.

Key Letta features:

• Context Repositories — Git-based memory for coding agents
• Conversations API — Shared memory across concurrent experiences
• Letta Code — #1 model-agnostic agent on Terminal-Bench
• Self-improvement — Agents that learn from their mistakes

Security

Strengths

• Memory-First — Persistent memory is the core feature, not an afterthought
• Unified Context — One agent across all channels
• Platform Backing — Built on mature Letta infrastructure
• Voice Support — Whisper transcription across all channels
• Skills Ecosystem — ClawHub + skills.sh compatibility
• Self-Hosted — Run on your own infrastructure

Weaknesses / Considerations

• Archived — No new development since May 2026; the repo is read-only and Letta points new users to Letta Code Channels instead^[1]
• Successor Requires Letta Cloud — Channels in Letta Code is cloud-only (not available in Local mode) and currently covers Telegram, Slack, Discord, and WhatsApp — Signal did not carry over^[2]
• Letta Dependency — Required a Letta API key or self-hosted server
• Never Reached Stable — All 15 releases were alphas (v0.1.0-alpha.1 through v0.2.0-alpha.13)

What Developers Say

There is little public commentary on LettaBot specifically — it was archived roughly four months after its first alpha, before a substantial user community formed. The most relevant discussion is the Hacker News thread on Letta Code, the product that absorbed it, where the debate centered on whether persistent agent memory helps or hurts:^[4]

"Maintaining the memory is a considerable burden" — tigranbs, Hacker News

"ChatGPTs implementation of Memory is terrible. It quickly fills up with useless garbage" — DrSiemer, Hacker News (arguing for Letta's more transparent, controllable approach by contrast)

Letta team members in the same thread emphasized that their white-box memory system is inspectable and editable, unlike opaque alternatives.

Who Built It

Letta AI, the company behind MemGPT. A UC Berkeley AI research lab spinout founded by Sarah Wooders and Charles Packer, Letta raised a $10M seed round led by Felicis with participation from Sunflower Capital and Essence VC, at a reported $70M post-money valuation.^[5] The team includes the researchers who pioneered the MemGPT concept.

Ideal User

• Existing LettaBot deployments — they continue to function, but plan a migration
• Memory-focused users should look at Letta Code with Channels, the official successor^[2]
• Letta ecosystem users already on Letta Code get the same multi-channel idea without running a separate bot

Bottom Line

LettaBot was the memory-first entry in the personal-agent wave: one agent, one unified memory, five messaging channels. That architecture was genuinely different from multi-channel bots that maintain separate contexts — and it proved compelling enough that Letta folded it into its flagship product. The standalone repo was archived in May 2026 at 328 stars, replaced by Channels and Schedules built directly into Letta Code.^[1]

Treat LettaBot itself as historical. If the pitch — an assistant that actually knows you over time, reachable from Telegram, Slack, Discord, or WhatsApp — still appeals, the path forward is Letta Code with Channels on Letta Cloud.^[2] The company behind it remains well-capitalized ($10M Felicis-led seed) and active, with the core Letta platform at 23k+ GitHub stars.^[5]

---