CIRIS is open-source infrastructure for trustworthy AI — ethics and intuition working together. Many bounded agents, each auditable, each accountable to local humans. Ally is your personal assistant built on it — runtime ethical governance, coherence monitoring, cryptographic audit trails.
CIRIS Android App Now Available
Images, documents, Home Assistant integration — your ethical AI, on mobile
Ethics alone isn't enough. There are three types of AI — and only one avoids the worst outcomes.
Unethical AI fails obvious tests. Ethical AI passes them but acts with false confidence — when correlated sources agree, it can't tell the difference. Ethical + Intuitive AI monitors its own coherence and knows when agreement is too easy. CIRIS implements all three layers: ethics, intuition, and the Coherence Ratchet.
Ally is an open-source AI assistant that runs on your hardware. Task management, scheduling, decision support, smart home control — with complete transparency. Core logic runs locally; LLM inference uses your choice of provider (cloud APIs, local models, or our zero-data-retention proxy). Every decision is auditable, every action is accountable.
Watch your agent's knowledge grow in real-time. Interactive SVG visualization shows how concepts connect, cluster, and evolve. Radial, force-directed, or hierarchical layouts. It's beautiful and it's yours.
Four operational modes — WORK, PLAY, SOLITUDE, DREAM — each with different learning behaviors. Behavioral drift monitoring ensures the agent stays aligned as its knowledge grows. Configurable thresholds trigger recalibration.
Each truthful action makes future truth-telling easier and coordinated deception harder. Honest actions reference prior signed commitments directly. Deceptive actions must satisfy an ever-growing constraint surface of immutable rationales, identity bounds, and observed outcomes—becoming increasingly fragile under scrutiny.
Auto-detects sensitive information — API keys, passwords, personal data. AES-256-GCM encryption. The agent self-configures new patterns as it learns what you consider private.
Intelligent message prioritization with spam detection and user trust tracking. The agent learns who matters and what's noise. Your attention is protected.
Artificial Interaction Reminders maintain clarity. After extended sessions or high message counts, the agent reminds users it's software. Configurable thresholds. Transparency by default.
Six reasoning events streamed live via SSE as the agent processes each thought. Watch the H3ERE pipeline in action — DMA analysis, action selection, conscience validation, execution. Nothing hidden. Explore a real trace →
Full OTLP export — metrics, traces, logs. Plug into Jaeger, Prometheus, Grafana, any observability stack. Cognitive processing traces with span IDs, thought steps as events, service attributes.
The ConsentImpactReport schema tracks interactions, patterns contributed, and users helped. Early infrastructure for understanding how value flows through a collaborative system.
Five-step alignment check before work begins. Verify configuration, validate system integrity, confirm service health, acknowledge operational limits, log initialization. Higher-stake deployments run the full sequence.
Shutdown requests flow through the agent's normal task processing. The agent can acknowledge, defer if mid-task, or flag concerns. Configurable modes from instant to fully-managed transitions.
Multiple instances coordinate system-level decisions through shared task ownership. One instance claims decisions for the group, with full audit trail. Scale horizontally with consistent behavior.
Get started with CIRIS in seconds
pip install ciris-agentWindows, macOS, and Linux supported | Python 3.10+
See all 5 installation methods →Before asking for help, Ally analyzes every decision through ethical, common-sense, and domain-specific lenses. Six conscience checks validate every action. If uncertainty remains, it escalates to you or advisors you trust. You maintain final authority.
Complex ethical questions get escalated to you or advisors you trust. CIRIS doesn't pretend to have all the answers — it knows when to ask for human wisdom.
Every decision runs through three parallel analyses: ethical principles, common-sense plausibility, and your specific context. Then six conscience checks validate the action before execution.
Core logic, database, memory, and audit logs run locally. LLM inference uses your choice of provider — cloud APIs, local models, or our zero-retention proxy. No data harvesting, no surveillance.
Built on The Three Rules: No Untyped Dicts, No Bypass Patterns, No Exceptions. Every interaction is validated through Pydantic models, creating an immutable audit trail.
From ethical reasoning and memory management to secure communications and audit logging—CIRIS coordinates specialized services that work together transparently.
Inspect the code, verify the behavior, contribute improvements. Transparency isn't a feature—it's the foundation of trust.
Ally is just the beginning. The entire CIRIS stack is open source — the agent framework, the zero-data-retention proxy, the billing system, everything. Because AI that serves humanity should be transparent, auditable, and owned by everyone.
Let's keep in touch! Sign up to receive updates and be the first to know when we launch.