The interface. Clean, persistent, yours.
AI tools answer questions. Chalie helps you think and act.
Modern tools are fast, but forgetful. Conversations reset — context is lost between sessions. Notes accumulate without meaning. Automation acts without awareness.
This creates quiet friction: repeating yourself, losing threads of thought, missing connections, acting without full awareness.
Chalie exists to reduce that friction. Not by replacing judgment — by staying alongside it.
The interface between you and the digital world.
Chalie is a cognitive ecosystem that earns trust over time. Today it thinks alongside you — remembering context, reasoning about intent, and acting through a growing set of tools. The long-term goal is a single place through which you interact with services, tools, and technology.
Chalie never acts without your approval.
- • Persistent memory across sessions — episodes, facts, traits, concepts
- • Research, search, and information retrieval via 15 tools
- • Gmail, Google Calendar, and Telegram integration
- • Multi-session background tasks with plan decomposition
- • Broader tool ecosystem — email, messaging, smart home, commerce
- • Graduated autonomy — spending tiers, scoped permissions, trust calibration
- • Single interface for all digital intent
Chalie is 100% open source. Anyone can inspect, contribute, and extend it. Tools integrate via containerised modules. Built in the open so trust can be verified — not assumed.
What guides the design of Chalie.
Context that persists
Conversations reset — context is lost between sessions. Chalie maintains continuity so you do not have to carry everything alone.
Memory that understands
Not search results. Episodic memory, facts, concepts, and traits — layered the way human cognition actually works.
Local first, private by design
Your data stays on your machine unless you choose otherwise. No telemetry. No external calls unless you configure them.
Model agnostic
Works with Ollama, OpenAI, Anthropic, and Gemini. Swap providers without losing continuity.
It acts, but does not take control
Chalie suggests actions when the moment calls for it. It does not automate your life. It stays alongside.
Think
Clarifies when something is missing. Asks rather than assumes. Routes to the right mode — respond, clarify, act, or simply acknowledge.
Remember
Working memory, gists, facts, episodes, concepts. Six layers of context that operate on different timescales and integrate automatically.
Act
When the moment calls for it, Chalie can search, retrieve, and take actions via containerized tools — with hard safety limits.
Step 1: Clone
git clone https://github.com/chalie-ai/chalie.git
cd chalieStep 2: Start
docker-compose build
docker-compose up -d
# Open http://localhost:8081/on-boarding/What does Alpha mean?
Chalie is in active early development. The core ideas are proven and the system runs — but you should expect rough edges, breaking changes, and missing features. Here's what that means in practice:
APIs, data schemas, and config formats may break between versions without a migration path.
Some flows are incomplete or unreliable. We ship fast and fix fast — issues are expected and welcomed.
You can follow every decision in the build log. Early users shape the direction of the project.
Found a bug or have a suggestion? The project is fully open source.
Web Search
Search the web for current information
Calendar
Check and manage calendar events
Reminders
Set and retrieve scheduled reminders
More in Embodiments →
Feb 27, 2026
Cognitive Reflexes and Structured Task Planning
A major update introducing learned 'reflexes' to bypass the main cognitive loop for common queries, alongside a new system for decomposing complex tasks into structured plans.
Feb 26, 2026
Zero-Config Voice, Cognitive Legibility, and a Testing Blitz
Shipped a new self-contained voice service, made Chalie's internal thinking visible through UI updates, and massively expanded the project's test suite to nearly 1,000 tests.
Feb 25, 2026
Architecting Awareness and Fixing a Broken Build
Shipped several major new architectural components for ambient awareness and self-correction, while also fixing a cascade of silent failures in the tool profile build system.