1. How Oli Works: Architecture and Intelligence
Oli is powered by a multi-agent system, where each agent plays a specific role from open-ended dialogue and emotional tone detection to guiding users through goals, activities, and reflection prompts. Rather than relying on a single monolithic model, Oli’s architecture is modular and designed for adaptability.
At its core, Oli leverages cutting-edge LLMs (currently Anthropic’s Claude) for natural conversation, surrounded by a system of specialized agents that:
- Interpret emotional tone
- Detect high-risk situations and escalate to human support if needed
- Provide structured guidance for coaching or well-being
- Manage and update dynamic memory (with consent)
This structure allows Oli to respond contextually, adapt its tone and suggestions, and deliver a tailored experience while remaining scalable and efficient.
2. Key Features for Users and Partners
- Dynamic Memory
Oli remembers key information over time such as goals, past challenges, and emotional cues to create a sense of continuity and depth. This memory is designed to fade or consolidate over time, mimicking human memory patterns. - Structured Pathways
Users can engage in specific journeys such as improving focus, managing stress, or preparing for performance reviews guided by conversational prompts and micro-interventions. - Adaptive Tone and Format
Oli adjusts to each user’s style, whether they prefer casual conversation, structured guidance, or short interactions. - Human-in-the-Loop Support
If a user’s needs go beyond Oli’s scope, the system automatically recommends escalation to a human therapist or coach, when available. - Analytics for Managers (Optional)
For enterprise customers, anonymized insights can be provided at an aggregate level to help leaders understand trends and team needs without compromising individual confidentiality.
3. Privacy, Data Control, and Safety
We’ve implemented a privacy-first approach to Oli’s design, ensuring it meets both user expectations and enterprise requirements for data security and confidentiality.
- Opt-in Memory
Users are informed clearly, and they can review delete threads from Oli which will remove what was discussed from the memory . When memory is off (incognito conversations coming soon), no information is retained between sessions. - End-to-End Encryption
All user data is encrypted both at rest and in transit. Memory data is stored separately and securely, with strict access controls. - No Employer Access to Conversations
Individual conversations remain strictly confidential. Employers and administrators never have access to personal data or transcripts. Optional insights available to organizations are fully anonymized and always aggregated. - Crisis Detection and Escalation
Oli includes dedicated agents to detect language that may indicate emotional distress or urgent risk. These agents respond by redirecting users to human professionals or appropriate resources, without attempting to handle crisis scenarios directly. - Provider Compliance and Auditing
We have signed agreements with all AI third-party providers used in Oli’s infrastructure to ensure they’re HIPPA-compliant (Data Infrastructure & Processing Assurance) and have zero data retention in their services. This ensures that all vendors involved in the processing of user data meet strict standards of security, confidentiality, and lawful processing including requirements around storage, access control, and auditing.
4. Optional Data Use for Product Improvement
In addition to protecting user privacy, we also believe in transparency about how we improve Oli. Users have the option to explicitly opt in to allow their conversations to be used for product development and enhancement.
Importantly:
- No human reads conversations directly.
All analysis is conducted by AI agents trained to identify patterns, detect usability issues, and improve performance across conversation flows. - Purpose-limited use only.
These conversations are never used for marketing, shared with third parties, or exposed to employers. They are only used to improve how Oli understands user needs, responds empathetically, and delivers value. - Opt-in, opt-out at any time.
Participation is entirely voluntary and can be withdrawn at any time from within the user settings. Once disabled, all past data marked for product improvement is removed from analysis. - Aggregated insight only.
The system only surfaces non-identifiable insights (e.g. topics that need better handling, tone mismatches, or UX friction), which help us fine-tune Oli’s responses and identify areas for improvement.
This approach allows us to continuously improve Oli while keeping user data private, anonymized, and securely handled through automated means.
5. Designed for Scalability and Trust
Oli was built to meet the needs of growing organizations looking for reliable, cost-effective, and scalable support solutions. Its architecture ensures high availability, low latency, and consistent performance across devices and platforms.
As we continue to expand Oli’s capabilities, we remain committed to transparency, user agency, and continuous improvement through feedback and data always with the end goal of helping people grow.
Photo by Gertrūda Valasevičiūtė on Unsplash