A
ADE
PLATFORM
ON
← HOME

Platform — ADE Brain · The local memory layer powering all ADE infrastructure.

ADE
⬡
Memory Layer · Core Infrastructure · Always On
ADE Brain

The persistent memory layer that powers all of ADE. Brain stores episodic and semantic memory across every module and agent — running as a Windows service, always local, always available. The foundation without which no ADE tool reaches its potential.

● Live · Service
ADE BRAIN
Observer · Orchestrator · Memory · Policy · Executor
Four SQLite databases with WAL mode
FTS5 multi-shard oracle search across all DBs
Eight memory tools exposed to all agents
LLM distiller — episodic → semantic facts
Two-level Observer (filesystem + pipeline)
JSONL pipeline monitoring on all REPLs
FastAPI REST on port 8766
Windows service via NSSM, ADE_ROOT env
Working · Episodic · Semantic memory levels
brain_client.py for any module integration
Prompt caching — 3-tier static / semi-static / dynamic
ADE Registry on file-based module manifest
3
Memory
Levels
7
SQLite
Databases
21
MCP
Tools
8+
Modules
Patched
Modules Patched with Brain Episodic

claude_repl.py ✓  ·  gpt_repl.py ✓  ·  3flash.py ✓  ·  ade.py ✓  ·  ADE Lite engine.py ✓  ·  AutoTask Swarm orchestrator.py ✓  ·  CodeSwarm server.py ✓  ·  Browser Agent server.py ✓
Pending: GigaMail · ADE CRM

⬡
ADE Brain — Connected
CodeSwarm stores failure patterns, repair strategies, and test results in Brain for compound learning
Failure MemoryPattern LearningRepair History
⌥
Autonomous Dev Pipeline · 5 Agents
CodeSwarm

Five-agent autonomous development pipeline where each model plays a specialized role — one repairs, one approves, one has the final word, each reviewing the previous one's output. The same pipeline used to build ADE itself. Planned open-source release as ADE Analyzer.

● Live · Internal
Pipeline Architecture
Claude
Orchestrator
GPT
Analyzer
Gemini
Repairer
Claude
Approver
GPT
Tester
5-agent review pipeline
Mechanical fix — Approver → snippet pairs → Repairer
Git rollback on repair failure
Swarm memory with FTS5 + BM25
Failure classification and tagging
Level 2 pattern generalization
Level 3 roadmap template distillation
REST API + benchmark dashboard

Open Source Strategy: CodeSwarm / ADE Analyzer planned as open source (BSL or AGPL) to drive developer adoption. ADE Core remains closed source — it is the structural moat. Same playbook as HashiCorp, Elastic, MongoDB.

⬡
ADE Brain — Connected
AutoTask records every execution, failure, and repair into Brain for compound learning and pattern distillation
Execution LogCompound Learning
⚙
Autonomous Task Execution · Multi-Agent
AutoTask Swarm

Multi-agent task execution framework with explicit build graphs, rollback on failure, and compound learning. AutoTask breaks complex projects into steps, assigns them to the right agent, monitors execution, and learns from failures. The backbone of ADE's autonomous execution capability.

● Live · Production
Explicit input_files / verification_targets
rollback_surface via git_rollback.py
Swarm memory with FTS5 + BM25
Failure classification and structured tagging
Level 2 — pattern generalization
Level 3 — roadmap template distillation
Benchmark suite with REST API + dashboard
Semantic verification via integration_tester.py
15/15
Steps First
Full Run
100%
Repair Success
(Internal)
3
Learning
Levels
5
Architecture
Improvements
⬡
ADE Brain — Connected
ADE Lite logs every step execution and output to Brain for cross-session continuity
Step MemorySession History
⬡
Lightweight Execution Engine · FastAPI · Port 8009
ADE Lite

Stripped-down version of the ADE execution engine for rapid prototyping and focused task execution. Runs roadmaps with write_file, run_command, and run_python steps — no overhead, no hallucinations. Used to build ADE's own vertical modules — a full vertical was built end-to-end in a single session: 40 steps, zero manual intervention.

● Live · Production
Roadmap-driven execution (YAML / JSON)
Three step types: write_file · run_command · run_python
Dynamic max_tokens by task type
Auto-exec for read-only commands
Anti-loop safeguards built-in
Confidence scoring per step
ADE Brain integration for memory
ADE Registry module system
Executor Agent — tool-calling logic separated from REPLs
Relative paths via ../brain/executor/ from backend/
40
Steps
Full Module Build
1
Session
Zero Interventions
8009
FastAPI
Port
3
Step
Types
⬡
ADE Brain — Connected
TestSwarm logs all test results and regressions to Brain — failures inform future CodeSwarm runs
Test HistoryRegression Memory
🔬
Autonomous Product Tester · Integrated Pipeline
TestSwarm

Autonomous product testing integrated into the CodeSwarm pipeline. Runs the Tester agent role — verifying that repairs actually solve the problem and don't introduce regressions. Three-state verification: passed, failed, or skipped with reason. All results feed back into swarm memory.

● Live · Integrated
Three-state verification — passed / failed / skipped
Regression detection after each repair
tester_unit.py — pytest for pure Python
tester_api.py — httpx for FastAPI endpoints
Integration test runner with route validation
Benchmark API with performance dashboard
Test results feed into swarm memory
Pattern learning from repeated failure classes
3
Verification
States
2
Tester
Modules
0
Regressions
First Full Run
∞
Learns From
Every Run
● Idle