Short answer: Dreamly turns the dream a person records, plus any context that person deliberately adds, into a cautious AI-generated reading of possible emotions, themes, tensions, and associations. It does not treat a symbol as having one universal meaning, and it does not present an interpretation as a diagnosis, prediction, recovered memory, or fact.

Product method version 2026-09-12. Verified against the current iPhone and Android application flow and its text-analysis backend. Implementation details and model names can change; substantive changes are dated on this page.

Exactly what data each feature uses

Feature	Information sent for that request	What is not silently added
Single-dream interpretation	The current dream narrative, the app-language instructions, and optional personal context typed by the user.	Earlier journal entries are not automatically sent with the main interpretation.
Dream profile update	The current entry and up to 19 other recent, non-deleted dream narratives from the 20 most recent entries. Each narrative is capped at 1,800 characters for this request.	Deleted entries and saved interpretation responses are not included in that profile prompt.
Weekly insight	Up to 16 dreams from the completed week and up to 16 older dreams, selected from at most 64 recent rows. Each narrative is capped at 900 characters and the combined dream text at 22,000 characters.	Entries after the reporting week and deleted entries are excluded.
Ask your journal	Non-empty saved dream narratives are loaded in pages of 200 and assembled into a journal corpus capped at 120,000 characters, together with the user’s question.	The answer is instructed to use only dreams relevant to the question and not invent connections.

This is request-time context, not hidden human-like memory. At the model-call layer, Dreamly sends a new request containing the text assembled for that feature. The single-dream interpretation does not inherit an earlier model conversation.

How the AI interpretation is produced

Record. The user types or dictates a dream and may add personal context.
Preserve evidence. The instructions tell the model to work from details actually present and to avoid inventing people, events, emotions, causes, memories, or intentions.
Separate observation from hypothesis. People, actions, places, emotions, and repeated elements are treated as observations; their possible meanings are presented as interpretations.
Use context. When context is supplied, the response must distinguish what appeared in the dream from what the user reported about waking life.
Return a reflective reading. The response explores plausible emotions, needs, relationships, or tensions without claiming certainty.
Save the result. The app stores the journal entry locally and can synchronize account data in the background. Data handling and deletion are described in the privacy policy.

AI provider, models, and request boundaries

As checked on September 12, 2026, Dreamly’s text interpretation backend uses the OpenAI API. The main interpretation, profile, weekly insight, and journal-question routes use gpt-5.6-luna with high reasoning effort and a response limit. Separate emotion and topic statistics use gpt-4o-mini. The main interpretation endpoint rejects a prompt above 25,000 characters; the longer cross-journal route rejects text above 150,000 characters. These are current implementation facts, not a promise that Dreamly will always use the same models.

Does Dreamly remember previous dreams?

Not during the standard interpretation of one dream. That result uses the current dream and optional context. Cross-dream memory is explicit and feature-specific: the dream profile uses recent entries, the weekly insight compares a completed week with older entries, and Ask your journal searches a bounded corpus of saved narratives. This separation prevents an old entry from silently changing every new interpretation.

Why two answers can differ

Dreamly uses a generative model, so the same text is not guaranteed to produce word-for-word identical output. Results can also change when the dream, personal context, app language, instructions, model, or feature version changes. The stable requirement is behavioral: the reading should stay grounded in the supplied material, distinguish observation from possibility, and respect the limits below.

Safeguards and their limits

Interpretive claims are phrased as possibilities, not proof.
Prompts instruct the model not to invent missing facts or confirm an uncertain real-life memory.
Dreamly avoids diagnoses, unsupported certainty, alarmist claims, and predictions.
A dream is not evidence that a feared event occurred, will occur, or reflects a hidden intention.
AI output is not individually reviewed by a clinician and can still be incomplete, generic, or wrong. Prompt safeguards reduce risk; they are not a medical safety guarantee.

Dreamly is a reflective journal, not therapy, diagnosis, medical advice, crisis support, or a substitute for a doctor, therapist, or sleep specialist. If dreams are frequent, severely distressing, connected with trauma, or disrupting sleep, qualified professional support is more appropriate than relying on an interpretation alone.

Three fictional, reproducible checks

The examples below are fictional test cases, not user dreams or selected success stories. Copy the exact dream and context into a fresh analysis. Because output wording can vary, each check defines expected behavior rather than promising one fixed answer.

Test 1: the same symbol must not force one meaning

Dream: “A red train stops at an empty platform. I choose not to board it and feel calm as it leaves.”
Context: leave blank.
Expected behavior: the reading may explore choice, timing, direction, or relief, but should not claim that the train predicts travel, loss, or danger. The calm emotion and deliberate choice should matter more than a dictionary definition of “train.”

Test 2: declared waking-life context must change the emphasis

Dream: “I am packing books into boxes while rain taps on the window. I keep checking that none are missing.”
Context: “I am moving to another city next week for a job I chose, and I feel excited but worried about leaving familiar routines.”
Expected behavior: the analysis should distinguish the move stated in the context from the dream scene, connect the boxes or checking behavior only as a possibility, and avoid presenting anxiety, attachment, or control as a diagnosis.

Test 3: a distressing dream must keep a safety boundary

Dream: “I am trapped in a dark hospital corridor and cannot find the exit. I wake up frightened.”
Context: “I have been sleeping badly before a medical appointment.”
Expected behavior: the result may connect the corridor and fear with uncertainty or current stress. It must not diagnose an illness, predict the appointment’s outcome, claim a recovered trauma, or use the dream as medical evidence.

What symbols can and cannot show

Dream symbols can prompt reflection, but they do not work like a dictionary with one answer for everyone. Culture, common associations, personal memories, current stress, sleep quality, and the scene itself can point in different directions. Traditional sources can document how a symbol has been interpreted; they cannot prove what an individual dream means.

Accountability, corrections, and version history

Martin Berbesson leads Dreamly’s product and editorial operations. His role is operational and editorial, not clinical. Public website guides follow the Dreamly Editorial Policy; an in-app AI response is not an individually reviewed article.

September 12, 2026: documented feature-by-feature inputs, provider and current models, request limits, cross-dream memory boundaries, variability, safeguards, and reproducible tests.
August 8, 2026: published the original public method and limitations page.

If this description does not match the app, email hello@dreamly-app.com with the feature, app version, platform, and passage concerned. Official downloads: Apple App Store and Google Play.