CIRIS is open-source accountability infrastructure for autonomous AI — coherence and conscience working together. Many bounded agents, each auditable, each accountable to local humans. Ally is your personal assistant built on it — runtime conscience, coherence monitoring, cryptographic attestation.
CIRIS Mobile Apps Now Available
Images, documents, Home Assistant integration — your auditable AI, on mobile
EthicsEngine — Measuring Ethical Reasoning in LLMs
An affiliated project for evaluating the ethical reasoning quality of large language models. Transparent scoring, cryptographic audit trails, open to everyone.
Rules alone aren't enough. There are three types of AI — and only one avoids the worst outcomes.
Unaccountable AI fails obvious tests. Rules-Only AI passes them but acts with false confidence — when correlated sources agree, it can't tell the difference. Rules + Awareness AI monitors its own coherence and knows when agreement is too easy. CIRIS implements all three layers: conscience, 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.
Control your lights, thermostats, and devices with natural language. 'Turn off the living room lights' or 'Set the temperature to 72°.' Full Home Assistant API support — your smart home, your way, with full audit trails.
Get directions, traffic updates, and route planning through natural conversation. 'How do I get to the airport?' or 'What's the fastest route home?' Local processing means your location history stays private.
Real-time weather forecasts, severe weather alerts, and planning assistance. 'Should I bring an umbrella tomorrow?' or 'What's the best day to mow the lawn this week?' Context-aware answers that actually help.
Upload PDFs, images, and documents. Ask questions about your files — receipts, contracts, manuals. 'What's the return policy in this receipt?' Everything stays local unless you choose otherwise.
Create reminders, manage your schedule, track to-dos. 'Remind me to call Mom on Sunday' or 'What's on my calendar this week?' Natural language interface to your digital life.
Ask anything — from quick facts to complex research. Unlike other assistants, Ally tells you when it's uncertain and can defer to web search or human experts when needed.
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 coherent 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 coherence 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. Four ethical evaluations plus a four-faculty conscience 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 the four-faculty conscience validates 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 signed 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 legibility.
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.