all the api requests. pretty please.

1. Ecosystem Context & Impact

Target Audience: Third-party developers building integrations (Browser Extensions, Desktop Clients, Meeting Bots, IDE Plugins).
Current Limitation: The Developer API is currently restricted to passive, unformatted data dumping.
The Goal: Unlocking these features will transform Omi from a standalone wearable into a centralized AI hub that seamlessly connects with a user's entire digital workflow.

Problem: Injecting a structured transcript_segments array via the API does not trigger the native, colored chat bubble UI in the mobile app.
Symptom 1: Sending only transcript_segments results in a 422 Unprocessable Entity error.
Symptom 2: Sending both transcript_segments and the mandatory flat text field results in a 200 OK, but the app defaults to rendering the flat string as a single, dense note. The "Transcript" tab disappears entirely.
Broad Advantage: Developers need the native Transcript UI to format meeting transcripts, long articles, or desktop interactions. Forcing flat text creates unreadable walls of text, severely degrading accessibility for users with visual impairments or cognitive load constraints (e.g., ADHD/Dyslexia).
Request: Modify the API schema to natively accept and render segment arrays from 3rd-party sources without requiring flat text fallbacks that break the UI.

Problem: Omi's backend summarization LLM crashes when processing specific industry terminology.
Symptom: Pushing text containing specific terms (e.g., cybersecurity documentation, medical research regarding diseases, or creative fiction) results in a silent "Error while processing conversation" failure.
Broad Advantage: Users must be able to ingest their professional and educational literature without the AI falsely flagging it as a safety violation.
Request: Implement a developer flag in the POST request (e.g., safety_filter_level: "low"), tune the backend LLM system prompt, or introduce an age verification gate in the user profile that allows adult users to opt into non-standard, unfiltered LLM models for professional use.

Problem: Omi relies entirely on AI semantic analysis to route conversations into folders (Work, Personal, Social).
Symptom: High-token payloads (like a 2,000-character web article) dilute any NLP prompt injections or explicit category tags, causing the AI to miscategorize the data.
Broad Advantage: Reliable categorization is a fundamental requirement for 3rd-party apps. If a developer builds a "Fitness Tracker" integration, the data must go to a "Fitness" folder. Relying on AI guessing breaks app structured organization.
Request: Add a definitive folder_id or category_override string parameter to the API POST schema to bypass AI semantic sorting entirely.

Problem: Current endpoints (/memories, /conversations) are strictly one-way data ingestion pipelines.
Symptom: Developers can push context to Omi, but the API does not return the AI's response or summary in the HTTP reply.
Broad Advantage: Exposing a two-way endpoint unlocks a massive ecosystem of interactive tools. Users could query their Omi database directly from a desktop application, a browser extension, or a voice assistant, rather than being forced to open the mobile app for every interaction.
Request: Expose a /chat endpoint. Allow 3rd-party applications to pass context, send a prompt, and receive the LLM's response in the same synchronous HTTP transaction.