HOME / FAQ
Frequently asked questions
Short, direct answers to the questions engineers and compliance teams ask us most about Humane AI's behavioral intelligence layer.
What is Humane AI?
- Humane AI is a behavioral intelligence layer that sits between your application and any Large Language Model (LLM). It adds persistent memory across sessions, relational context per end-user, deterministic safety gates, proactive triggers, and A/B experiment infrastructure — accessible through a single API call across more than 20 LLM providers including OpenAI, Anthropic, Google Gemini, and AWS Bedrock.
How does Humane AI give LLMs memory?
- Humane AI stores behavioral signals — mood, energy, trust, familiarity, sentiment — as structured, retrievable state per end-user. When you send a new message, Humane retrieves the relevant behavioral context, combines it with semantic memory (MemPalace), and injects it into the LLM prompt. Memory persists across hundreds of sessions and respects per-plan retention caps for compliance.
Which LLM providers does Humane AI support?
- Humane AI supports more than 20 providers out of the box: OpenAI (GPT-4, GPT-4o, GPT-5), Anthropic (Claude 3.5 Sonnet, Claude Opus, Haiku), Google Gemini, AWS Bedrock (Claude, Llama, Titan), Mistral, Groq, xAI Grok, Together, Fireworks, Cohere, Hugging Face, Perplexity, local Ollama models, and more. Switch providers with a single string — behavioral context travels with the request.
Is Humane AI HIPAA-compliant?
- Humane AI is built for regulated industries. Per-plan retention caps, configurable PII redaction, audit logs, idempotency keys on every webhook, and self-hosted deployment options give healthcare and finance teams the controls they need. Our Clinic and Enterprise tiers include HIPAA Business Associate Agreements and audit-ready exports.
How do Humane AI's safety gates work?
- Safety gates are deterministic classifiers that run before and after the LLM call. The Values Boundary engine hard-blocks specific content categories using explainable matched-phrase audit. The Social Risk Engine produces graduated risk scores for sensitive interactions. Unlike prompt-based filtering, these gates return consistent results — the same input always produces the same gate decision, which auditors and compliance officers require.
Can I self-host Humane AI?
- Yes. The Community tier includes the core engines as self-hosted open-source software. The platform runs on any Docker-compatible host — we ship a production-ready Docker Compose bundle with PostgreSQL, Caddy (auto-HTTPS), and optional profiles for Ambuja demo and local Ollama. Paid tiers add the managed cloud dashboard, A/B testing, advanced analytics, and enterprise features.
How is Humane AI different from LangChain or LlamaIndex?
- LangChain and LlamaIndex are orchestration frameworks — they let you compose prompts, chain calls, and retrieve documents. Humane AI is a behavioral state layer — it tracks who the user is, how they're feeling, and what tone/timing/safety is appropriate right now. You can use Humane AI alongside LangChain: Humane provides the behavioral context, LangChain composes the final prompt.
How much does Humane AI cost?
- Humane AI has four tiers: Community (free, self-hosted), Starter ($49/month, up to 500 active users), Growth ($149/month, up to 5,000 active users, includes A/B testing), and Scale ($499/month, up to 50,000 active users, full engine suite). Enterprise contracts are custom and include SSO, SLA, and dedicated support. Annual billing saves 20%. No credit card required for the free tier.
What are proactive triggers?
- Proactive triggers let the engine reach out first — without a user message — when behavioral state crosses thresholds you define. For example: mood drops below 0.3 for three sessions, trust rises above 0.8, or a task reminder comes due. Triggers fire webhooks you handle, so the LLM can send a follow-up message, a wellness check, or escalate to a human.
Does Humane AI support A/B testing?
- Yes. The Growth and Scale tiers include a deterministic experiment framework with variant routing, exposure tracking, and two-proportion z-test analytics. You can test different tones, different safety thresholds, different memory retention windows — anything controlled by Humane's policy engine — and measure impact on completion rate, retention, or a custom metric.
How long does integration take?
- The Python and TypeScript quickstarts take about five minutes end-to-end. The SDK wraps your existing LLM client — you replace one import and one line to enable behavioral intelligence on every subsequent call. The cURL quickstart is three minutes if you want to integrate from another language directly over HTTP.
What data does Humane AI store?
- Humane AI stores the messages you explicitly send to it, derived behavioral signals (mood, energy, trust — never the raw message unless you opt in), and minimal audit metadata for compliance. Retention windows are configurable per plan and per end-user. Messages are encrypted at rest and in transit, and you can hard-delete any end-user's data via API or dashboard at any time.
STILL HAVE QUESTIONS?
Talk to us.
Engineers and product teams get a reply within one business day. Email hello@humaneai.vaarak.com or start the Python quickstart.