Dreamline¶
Dreamline is a self-developed AI memory service by Techlogia. It serves as long-term memory for AI assistants like Claude Code and OpenAI Codex – automatic, self-learning and usable across projects.
The Problem¶
AI assistants forget everything after each session. Decisions, context knowledge and learned preferences are lost. Teams constantly repeat themselves and waste time explaining things the assistant has already been told.
The Solution¶
Dreamline automatically captures AI sessions, extracts relevant knowledge and consolidates it into a compact, self-updating knowledge store. In the next conversation, the assistant knows the project context, team preferences and past decisions.
graph LR
A[AI Session] --> B[Capture Session]
B --> C[Extract Knowledge]
C --> D[Consolidate]
D --> E[Update Memory]
E --> F[Next session benefits]
How It Works¶
Dream Cycle¶
Dreamline goes through four phases to transform raw data into structured knowledge:
| Phase | Description |
|---|---|
| Orient | Assess current knowledge state |
| Gather | Collect and analyze relevant sessions |
| Consolidate | Merge new insights with existing knowledge |
| Prune | Clean up outdated or redundant entries |
This process runs automatically in the background – configurable by interval, minimum number of new sessions and other criteria.
Memory Types¶
Dreamline stores knowledge in four categories:
Information about users, roles and preferences.
- What role does the user have? (Developer, designer, manager)
- What experience do they bring?
- How should the assistant communicate?
Lessons learned from corrections and confirmations.
- What works well? What should be avoided?
- Preferred workflows and conventions
- Confirmed decisions
Project-specific context knowledge.
- Current goals and deadlines
- Architecture decisions and their reasoning
- Dependencies and constraints
Pointers to external resources and systems.
- Where are bugs tracked?
- Which dashboards are relevant?
- Where to find specific documentation?
Features¶
Multi-Tool Support¶
Dreamline works seamlessly with different AI assistants:
- Claude Code – Automatic capture via hook system
- OpenAI Codex – Capture via filesystem watcher
- Both simultaneously – Knowledge is consolidated across tools
Intelligent Gate System¶
A five-level control system prevents unnecessary processing:
- Activation – Feature toggle per project
- Time window – Minimum interval between consolidations
- Throttle – Maximum check frequency
- Sessions – Minimum number of new sessions as trigger
- Lock – Prevents parallel processing
Quick-Extract¶
Immediate fact extraction after each session – without waiting for the next dream cycle. Important insights are available right away.
Smart Recall¶
AI-powered relevance search across all stored memories. Instead of simple keyword search, Smart Recall understands the context of a query and returns matching entries.
Web Dashboard¶
Management interface for all projects, provider selection and manual dream triggering – accessible at localhost:8100.
AI Providers¶
Dreamline supports various AI providers for processing:
| Provider | Description | Cost |
|---|---|---|
| Claude (CLI) | Claude Code with existing subscription | Included |
| Codex (CLI) | OpenAI Codex with Plus/Pro subscription | Included |
| Ollama | Local LLMs, usable offline | Free |
| Anthropic API | Direct API connection | Pay per use |
| OpenAI API | Direct API connection | Pay per use |
Technology¶
Dreamline is built on a lean, containerized stack:
- Backend: Python 3.12 with FastAPI (async)
- Database: PostgreSQL 16 with SQLAlchemy 2.0
- Scheduler: APScheduler for background processes
- Container: Docker + Docker Compose
- Migrations: Alembic
Installation¶
Prerequisites¶
- Docker and Docker Compose
- Claude Code and/or OpenAI Codex installed
- Optional: Ollama for local LLM processing
Quick Start¶
```bash
Clone the repository¶
git clone https://github.com/techlogia/dreamline.git cd dreamline
Configure¶
cp .env.example .env
Start¶
docker compose up -d ```
After starting, the dashboard is available at http://localhost:8100.
Configuration¶
Key settings in the .env file:
```ini
Dream intervals¶
AUTODREAM_ENABLED=true AUTODREAM_MIN_HOURS=12 AUTODREAM_MIN_SESSIONS=3
Quick-Extract¶
EXTRACT_EVERY_N_SESSIONS=1
AI provider (claude-abo, codex-sub, ollama, anthropic, openai)¶
DEFAULT_PROVIDER=claude-abo ```
API¶
Dreamline offers a complete REST API for integration and automation:
Key Endpoints¶
| Endpoint | Method | Description |
|---|---|---|
/api/v1/projects |
POST |
Create project |
/api/v1/sessions |
POST |
Record session |
/api/v1/dreams |
POST |
Trigger dream manually |
/api/v1/recall |
GET |
Search memories |
/api/v1/memories |
GET |
List all memories |
/api/v1/dreams/status |
GET |
Check dream status |
/api/v1/stats |
GET |
Usage statistics |
/health |
GET |
Health check |
Example: Record a Session¶
bash
curl -X POST http://localhost:8100/api/v1/sessions \
-H "Content-Type: application/json" \
-d '{
"project_id": "my-project",
"messages": [...],
"source_tool": "claude"
}'
Example: Search Memories¶
bash
curl "http://localhost:8100/api/v1/recall?query=deployment+workflow"
Support¶
- Documentation: This page and the README on GitHub
- Issues: Via the GitHub repository
- Contact: Via our contact form