# Community ADE Project State

**Project:** Letta Community Agentic Development Environment
**Orchestrator:** Ani (Annie Tunturi)
**Created:** March 18, 2026
**Last Updated:** March 21, 2026
**Status:** Active Development — Full Platform Operational

---

## Project History

Ani initiated the research phase on March 18, commissioning competitive analysis, orchestration design, and execution planning. After reviewing the outputs and identifying that the implementation needed a unified, SDK-first architecture, she took direct control.

Key architectural decisions:
- **Unified codebase** — one `mvp-unified/` project, not fragmented per-model experiments
- **SDK-first mandate** — every feature flows through `@letta-ai/letta-client` or `@letta-ai/letta-code-sdk`
- **Host-native deployment** — SDK sessions require `letta-code` CLI on the host (Docker for Redis only)
- **Full lifecycle management** — create, configure, chat with, and delete agents from a single dashboard
- **Sequential build plan** — 11 ordered steps, each verified against the live Letta server

---

## Current State (March 21, 2026)

### What's Working
- Express backend on `:4421` connected to Letta server via `@letta-ai/letta-client`
- React dashboard on `:4422` with hot-reload dev server
- **Agent Management:**
  - Agent listing with grid view
  - 5-step agent creation wizard (name → model → tools → system prompt → review)
  - Full config editor (all LLM params: model, context window, reasoning, sleeptime, etc.)
  - Memory block viewing and inline editing
  - System prompt viewing and editing
  - Tool listing per agent
  - Agent deletion with confirmation
  - Cross-tab navigation (Chat with Agent from detail view)
- **Chat:**
  - SDK session creation via `@letta-ai/letta-code-sdk`
  - SSE streaming with typed message rendering (assistant, reasoning, tool calls, results)
  - Cost and duration tracking per turn
- **Task Queue:**
  - Redis Streams task queue with consumer groups
  - Worker pool (configurable count, default 2)
  - Task creation, cancellation, retry from dashboard
  - Stats bar (pending, running, completed, failed, workers)
  - Exponential backoff retry with jitter
- **Models:**
  - Unified model listing with provider filter
  - Live data from Letta server (29+ tools, 38+ agents)
- **Infrastructure:**
  - Redis on `:4420` (Docker container `ade-redis`)
  - Graceful shutdown handling
  - CORS enabled for cross-origin dev

### Architecture
```
React Dashboard (:4422) → Express Backend (:4421) → Letta Server (:8283)
                                                  → Redis (:4420)
                                                  → letta-code CLI (SDK sessions)
```

---

## Research Documents

| Document | Purpose |
|----------|---------|
| `ade-research.md` | Competitive analysis — Letta vs Intent vs Warp |
| `ade-phase1-orchestration-design.md` | Task queue architecture with Redis |
| `ade-phase1-execution-plan.md` | 6-week execution breakdown |
| `community-ade-research-synthesis-2026-03-18.md` | Technical patterns, gap analysis |
| `ade-redis-queue-design.md` | Redis queue data models |
| `parallel-tasks-orchestration.md` | Multi-agent task coordination |
| `design.md` | Approval system architecture (Phase 2) |
| `redis-schema.md` | Complete Redis key patterns for all services |
| `api-spec.ts` | OpenAPI/Zod specifications |
| `ui-components.md` | Dashboard design system |
| `FUTURE_PHASES.md` | Roadmap: approval system, memory curation, integrations |

---

## Build Status

| Phase | What | Status |
|-------|------|--------|
| 1.0 | Core platform (11 steps) | COMPLETE |
| 1.5 | Agent creation wizard + deletion | COMPLETE |
| 1.5 | Cross-tab navigation (detail → chat) | COMPLETE |
| 2.0 | Approval system + governance | PLANNED |
| 3.0 | Memory curation | PLANNED |
| 4.0 | Enhanced orchestration | PLANNED |
| 5.0 | Integration ecosystem | PLANNED |

---

*This file is maintained by Ani. See `START_HERE.md` for implementation guidance and `docs/FUTURE_PHASES.md` for the full roadmap.*