The gate
A hardwired safety gate reads every message before anything else. It cannot be bypassed — and in a crisis, Mindlens answers from vetted templates only. No model improvises.
A wise friend that reads the feeling underneath your words, remembers what matters, and helps you build the tools to steady yourself — over and over, for life.
Free · Yours alone, not encrypted end-to-end · Not a replacement for professional care
The idea Mindlens is built on
“Therapy did not fix me. Therapy gave me the tools to fix myself — over and over again, for the rest of my life.”
Mindlens exists to help you build those tools: understand what you’re feeling, regulate it, and notice yourself getting better at it.
Every message moves through the same careful pipeline — safety first, always, with no way around it.
A hardwired safety gate reads every message before anything else. It cannot be bypassed — and in a crisis, Mindlens answers from vetted templates only. No model improvises.
A bounded graph reads the moment, recalls only relevant context, and chooses one careful response path. You see progress without being given a diagnosis.
Mindlens remembers what you've shared — people, deadlines, wins — and brings back only what this moment actually needs.
A team of agents drafts a reply in the voice of a wise coaching friend: practical, warm, and usually ending in a choice — breathing, journaling, or just talking. When distress runs high it drops the options and just stays with you.
Moods and progress are logged so next week's Mindlens knows what this week's you worked through — and the people and topics it picks up land in Memory, where you can change them.
/// The read
A bounded LangGraph workflow coordinates deterministic safety, hosted assessment, memory, retrieval, one specialist, and optional review.
Deterministic crisis screening runs before any hosted model call and can exit directly to controlled templates.
Groq returns a constrained assessment used for routing; it never diagnoses or presents emotional certainty.
The graph selects relevant preference signals from MongoDB without placing personal memories in the vector index.
Pinecone searches only the curated therapy corpus and remains optional when the service is unavailable.
One route writes the user-facing reply: empathy, mindfulness, journaling, reflection, routine, challenge, or distortion.
Complex or high-distress turns can receive one additional safety review; ordinary turns do not pay that cost.
Controlled crisis responses and ordinary replies are validated, persisted, streamed, and returned through the existing contract.
Generation runs on Groq — a smaller open model for simple turns, a larger one when the moment is emotional or complex. Crisis turns use zero LLM calls: vetted templates only.
/// One room, twelve weathers
Tap a state. The field, the shadows and every accent on this page crossfade over 1.6 seconds — nothing ever snaps.
Current read
Warm spectrum
/// Built in the open
/// Safety & privacy
Mindlens is support software. It is not a clinical service, a diagnostic tool, or an emergency service — and it is built so it can never quietly pretend to be one.
In immediate danger? Contact your local emergency services. Mindlens always surfaces real human resources first, before anything else.
Zero LLM calls when it matters most. Every crisis response is written and vetted by humans.
No agent, prompt or feature can skip it. Safety overrides convenience — by design.
Every database query scoped to you, rate limits everywhere, and never sold or used for advertising. Replies are written by Groq — this turn's message goes to them with emails, phone numbers and ID numbers stripped, though your name and anyone you've mentioned go with it, because that's what makes a reply sound like it knows you.
Your conversations are saved so next week picks up where this one left off. Anything Mindlens learns from them shows up in Memory first, where you can edit or delete it.
Mindlens names feelings in plain language and shows how sure it is. It never labels you.