Engineering Memory
for Software Projects.
DevMemory AI is an Engineering Memory Operating System for software projects. It continuously captures project knowledge, build history, engineering context, generated summaries, timelines, architecture understanding, and development artifacts to create long-term engineering memory.
AI forgets everything.
Developers pay the price.
Every AI coding session starts with amnesia. No memory of your architecture, your decisions, your dependencies, or your conventions. The result is wasted time, hallucinated code, and lost engineering context.
Thousands of wasted prompts
Every new session starts from zero. You re-explain your architecture, your conventions, your decisions — every single time.
AI hallucinates your codebase
Without real context, AI invents APIs that don't exist, imports from wrong paths, and suggests patterns you abandoned months ago.
Engineering context is lost
Why was this decision made? What depends on this module? What broke last release? That knowledge lives in someone's head — until it doesn't.
RAG isn't enough
Vector search retrieves text. It doesn't understand relationships, dependencies, or architecture. Your codebase isn't a document — it's a system.
Six stages from code
to engineering memory.
DevMemoryAI v1.1.0 continuously observes your repository, compiles AST symbols into builds, stores them in SQLite, and serves context via CLI and dashboard.
1. Initialize Repository (dmai init)
Run dmai init in your project root. DevMemoryAI creates the .devmemory directory, starts the Chokidar watcher, prepares the Engineering Compiler, and boots the local index.
$ dmai init
✔ Repository initialized
✔ Watcher started
✔ Engineering Compiler ready
✔ Engineering Index ready (.devmemory/index.db)
✔ Dashboard available at http://localhost:314152. Observe Changes (Watcher)
The Watcher continuously monitors source files, AST symbol modifications, and file creations without needing manual git triggers or cloud webhooks.
[Watcher] Listening to source changes
→ Event: file_modified src/runtime.ts
→ Event: file_modified src/compiler.ts
→ Transaction tx_042 opened (cooldown 3000ms)3. Group Events (Transaction Manager)
Raw file save events are grouped into logical transactions using adaptive cooldowns, preventing build churn from rapid intermediate keystrokes.
[Transaction Manager] Cooldown timer elapsed (3000ms)
→ Closing Transaction tx_042 (2 files changed)
→ Forwarding to Engineering Compiler4. Compile Build & Write to SQLite
The Engineering Compiler acts as the sole writer of Engineering Memory. It parses AST symbols and records an immutable build in .devmemory/index.db.
[Engineering Compiler] Processing tx_042
→ Extracted 12 nodes, 8 edges
→ Writing Build #3 (Commit f83c10a) to SQLite index
✔ .devmemory/index.db updated5. Query Memory via dmai ask
Developers and agents query engineering memory using dmai ask to retrieve instant project context from SQLite.
$ dmai ask "What is the project architecture?"
[Engineering Memory Answer]
Single Runtime orchestrating Watcher, Compiler & SQLite Index...
✔ Queried from .devmemory/index.db6. Embedded Dashboard (:31415)
Open http://localhost:31415 to inspect the interactive Cytoscape knowledge graph, build timeline, SQLite nodes, and runtime logs.
http://localhost:31415
→ Knowledge Graph · 142 nodes, 218 edges
→ Engineering Timeline · Build #1, Build #2
→ Embedded HTTP Server activeEverything your project knows,
connected.
DevMemoryAI doesn't just store files. It understands relationships — what depends on what, what changed when, what decisions were made, and what impact a change will have.
Central local orchestration runtime. Coordinates Watcher, Compiler, and SQLite Index.
One index. Every tool & developer.
Deterministic engineering context.
DevMemory AI v1.1.0 exposes your Engineering Memory through a fast single query engine. CLI queries (dmai ask, dmai summary) retrieve instant, structured context directly from SQLite.
# devmemory project manifest (.devmemory/project.json)
{
"version": "1.1.0",
"projectId": "proj_8f93a1c2",
"repository": {
"name": "devmemory-demo",
"path": "/workspace/devmemory-demo"
},
"ai": {
"provider": "ollama",
"model": "gpt-oss:120b-cloud",
"url": "http://localhost:11434"
},
"dashboard": {
"port": 31415,
"autoOpen": true
}
}Single Source of Truth
All nodes, edges, builds, and prompts are queried directly from local .devmemory/index.db with microsecond latency.
Zero Latency Queries
dmai ask returns architectural answers locally. No remote RAG bottlenecks, zero cloud dependencies.
Structured Context Output
Generate standardized Engineering Context summaries or structured JSON for CLI scripts.
Not another RAG layer.
Not another note app.
DevMemory AI is a fundamentally different approach. It builds a deterministic index over compiled AST transactions — not probabilistic similarity search over text chunks.
| Capability / Feature | DevMemory AIv1.1.0 Engine | Vector RAG | Markdown Notes | Copilot Context |
|---|---|---|---|---|
| Single-Runtime Architecture | ||||
| AST Engineering Compiler | ||||
| Local SQLite Index (.devmemory/index.db) | ||||
| Deterministic Knowledge Graph | ||||
| dmai CLI Command Suite | ||||
| Embedded Local Dashboard (:31415) | ||||
| Executive Summary & Context Generator | Partial | Partial | ||
| Engineering Timeline & Build History | ||||
| Works 100% Offline | ||||
| Zero Cloud & Zero Telemetry Dependency |
Works with your local
engineering stack.
DevMemoryAI v1.1.0 operates locally. Connects with all major AI code editors and Ollama local LLM models.
Give your project
permanent engineering memory.
DevMemory AI v1.1.0 source code is publicly viewable for inspection and evaluation. Follow the recommended setup below to build and initialize locally.