OpenCAS is a persistent autonomous agent with local state, durable memory, a web control plane, and provider-routed model access through open_llm_auth.
This release bundle reflects the codebase as it exists now. It does not describe an aspirational future package. If a command, endpoint, or UI surface is listed here, it should exist in the repo and in the running server.
- Persistent episodic and distilled memory backed by SQLite
- Provider-backed chat, voice, and embedding lanes via
open_llm_auth - Memory inspection, retrieval inspection, and connected atlas views
- Chat, operations, usage, daydream, identity, executive, schedule, platform, logs, and system dashboard surfaces
- Twilio-backed phone bridge support with owner and caller workspace separation
- Telegram pairing and chat integration
- Desktop-context Body Double observations with screenshot/OCR evidence, MPRIS media state, YouTube transcript retrieval, and optional local Whisper transcription of currently playing audio
- Background daydreaming, creative ladder promotion, and task orchestration
- Operator-facing audit, receipt, qualification, telemetry, and plugin-trust APIs
- OpenCAS keeps its state locally under the configured state directory.
- The default CLI state directory is
./.opencas. - Chat, voice, and embedding traffic normally goes to whichever provider/model you configure through
open_llm_auth. - The default embedding model is
google/embeddinggemma-300m. - Embeddings have a deterministic local fallback path when provider-backed embeddings are unavailable.
- The dashboard server defaults to
127.0.0.1:8080.
That means the project is local-state and operator-owned, but not "fully local" in the sense of requiring no external model providers.
git clone https://github.com/Es00bac/OpenCAS_runtime.git OpenCAS
cd OpenCAS
git clone https://github.com/Es00bac/OpenLLMAuth.git ../open_llm_auth
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python -m opencas --tuiThe TUI bootstrap configures provider material, model selection, and Telegram settings for the current repo state.
First boot is responsibility-gated. OpenCAS creates persistent continuity, not a disposable chat session. If you later delete the state directory, you delete that agent's continuity. The TUI asks you to acknowledge this before creation.
After configuration:
python -m opencas --with-serverFor a non-TUI fresh bootstrap, acknowledge that boundary explicitly:
python -m opencas --with-server --accept-bootstrap-responsibilityThen open:
http://127.0.0.1:8080/dashboard
The current dashboard includes these top-level tabs:
- Overview
- Health
- Chat
- Operations
- Schedule
- Usage
- Daydream
- Memory
- Identity
- Executive
- Platform
- System
- Logs
| Document | Purpose |
|---|---|
| Installation Guide | Accurate setup instructions for the current repo layout |
| Usage Guide | How to run OpenCAS and use its operator surfaces |
| Features | Product capabilities and subsystem summary |
| Key Terminology | Definitions for the OpenCAS vocabulary used in docs, code, and the dashboard |
| API Reference | HTTP and WebSocket surfaces exposed by the running server |
| Architecture | Runtime structure, loops, persistence, and subsystem boundaries |
| Changelog | Release notes for this documentation bundle |
| Release Website | Standalone release landing page |
This repo is not yet packaged as a PyPI install. The current requirements.txt expects the editable gateway dependency at:
../open_llm_auth/
The public gateway repo is:
https://github.com/Es00bac/OpenLLMAuth.git
If you move the repo to another machine or directory layout, update that dependency path or install open_llm_auth separately before running OpenCAS.
Before calling a release artifact accurate, verify these commands on the current code:
source .venv/bin/activate
python -m opencas --help
python -m opencas --with-server --accept-bootstrap-responsibility
pytest tests/test_dashboard_api.py -qOpenCAS is released under AGPL-3.0-or-later; see LICENSE.