Aletheia

Decision Intelligence (DI) Infrastructure by Actanomos.

Aletheia is the Decision Intelligence (DI) Infrastructure for engineering and operations teams. It automates and governs complex operational, risk, fraud, underwriting, and customer lifecycle decisions: customer onboarding, transaction monitoring, claims triage, spend approvals, refunds and returns, support ticket triage, and content moderation. Aletheia unites durable workflow orchestration, hybrid rule evaluation, multi-install AI models (ChatGPT, Claude, Gemini, and OpenAI-compatible endpoints), generic HTTP API connectors, an admin console for human-in-the-loop review and policy authoring, and dynamic collection journeys—governed as one platform with one verifiable audit trail.

Available as a cloud SaaS platform at actanomos.com or self-deployable in your private cloud/VPC.

What it is, and what it is not

Aletheia is Decision Intelligence Infrastructure. It gives teams the machinery to express decision policies (deterministic rules, rule sets, durable workflows, AI model prompts, collection journeys), execute them in real time, route edge cases to human reviewers when policy dictates, and maintain an immutable, verifiable audit trail of every input, rule calculation, and model inference. It ships with 20+ starter policy packs covering risk, fraud, spend, operations, and customer lifecycle decisions that you can customize, dry-run, and backtest before publishing.

Hybrid AI and deterministic rule execution

Aletheia bridges deterministic rule logic with modern generative AI reasoning:

• Deterministic rule engine: CEL expressions, score thresholds, regex matching, velocity windows, and list lookups provide strict, reproducible, sub-millisecond policy gating.
• AI model integrations: Model apps put language models directly into workflow steps (call_app) or rules (app rules). Models perform contextual reasoning, unstructured document analysis, and probability-scored decisions (decide) or narrative summaries (generate).
• Severity-gated safety: Model steps operate under severity policies (e.g. severity: 'warn'). AI models advise and explain—they never block unilaterally without human-in-the-loop escalation.

Generic multi-install architecture

Aletheia features a generic multi-install setup:

• AI models: Connect ChatGPT (OpenAI API or Azure OpenAI), Claude (Anthropic API or Amazon Bedrock), Gemini, and any OpenAI-compatible server (Groq, Mistral, Together AI, Ollama). A workspace can install multiple instances of any model app with dedicated endpoints, models, rate limits, and keys.
• HTTP APIs: The generic http-api app connects to any third-party REST API in Hoppscotch request format without writing code, with support for document attachments and signed webhook callbacks.
• Custom apps: Custom WebAssembly modules built with @aletheia-dev/app-sdk run in sandboxed isolation with egress host whitelisting and zero credential leakage.

Subsystems

• API (apps/api): High-throughput Fastify API with OpenAPI at /docs. Authenticates workspace users and machine tokens, validates requests against core schemas, initiates workflow runs, handles incoming webhooks, and records the immutable audit trail.
• Durable workflow engine & worker (apps/worker): Distributed workflow engine that interprets published policy definitions step by step. Pauses for dynamic collection submissions, handles multi-day reviewer waits with SLAs, invokes AI models, and dispatches outbound webhooks.
• Hybrid rule engine (packages/rule-engine): High-performance CEL expressions, score thresholds, velocity aggregation, list lookups, and AI app rules, combined under configurable severity policies.
• Apps and AI runner (packages/app-sdk, packages/app-host, apps/app-runner, packages/model-apps, catalog/*): Sandboxed execution runtime for multi-install AI models and vendor HTTP APIs. Enforces per-call WebAssembly sandboxes, egress host whitelisting, timeouts, retries, and secure credential injection where keys never touch worker logs or modules.
• Operations & review console (apps/admin-console): Collaborative case queues with SLAs, AI-generated case briefs, dual-control policy approvals (maker-checker), visual diffs, and live metrics.
• Dynamic collection terminal (apps/collection-terminal): Hosted or embedded user intake journeys with conditional branching, document verification, and signed expiring links.

Who it is for

• Engineering and operations leaders who need a unified Decision Intelligence platform rather than piecing together point tools for rules, LLMs, and case queues.
• Risk, fraud, and policy teams who need version-controlled policies with visual diffs, dry runs, and historical backtesting before publishing changes.
• Developers integrating AI models or vendor APIs: configure multi-install model apps in seconds or build custom WebAssembly apps with the published SDK.
• Analysts and reviewers who resolve escalated cases with full context, rule breakdown, and AI-generated narratives in front of them.

Where to go

You want to	Start with
Run it locally and see a decision happen	Get started, then Tutorial: a first decision
Build on it	Concepts, then Rules, Admin console: Define and Apps
Integrate AI models	AI models, App catalogue
Explore policy templates	Example policies: 20 starter packs across risk, fraud, customer triage, and spend
Deploy and operate	Run a pilot, then Deployment and Operations
Project and roadmap	Roadmap, then Changelog and Support