Skip to content

Feature suggestions

William M Schmidt edited this page Feb 12, 2025 · 4 revisions

A list of some larger suggested features the Jarvis team to look at.

Feature suggestions

Tree-like chat architecture

Be able to ask about certain parts of an answer, similar to how OpenWebUI does it. Generating a tree of knowledge in one chat/instance.

Deep research

Similar to OpenAI and Google's deep research or OpenDeepResearch.

Web exposure

Expose the Jarvis user interface to the web - requiring a username and password login, intrusion detection, exponential retries, hidden behind a CDN, and 2-factor authentication.

Obsidian/Personal files integration

Connect your obsidian folder so it gets indexed, hashed and continuously embedded and updated. EXTRA: Embedding files and hashtable should be inside the obsidian folder so when obsidian synchronizes it doesn't have to re-embed.

Actual reflection

Using an as or more powerful LLM, review a result or reasoning prompt in a chain. Have a different pre-prompt and review only the "conclusion" and nothing before. Then have the reviewer run its own chain-of-though to increase stability. Must be tested using benchmarks somehow.

Web research

Deep research feature-like https://github.com/nickscamara/open-deep-research Firecrawl

RAGAS

Evaluate LLM models https://docs.ragas.io/en/latest/

Data processing and JSON structuring

Get a screenshot of a meal-plan, recipe, workout or other data. Have it be processed and output as JSON with a specific schema for other applications to use.

Math engine

Wolfram alpha, symbollab and other graphing tools to help the model make accurate mathematical assessments when the user requires it to. Grounding it.

Electronic circuit diagram and simulation engine

To help prototype buck converters. LLM can already create reasonable specs, but can't help find and source parts within a reasonable budget (requires web) and can't actually test its assumptions. Layout is also super-hard.

Actual web browsing - Puppeteer integration

To let an LLM solve an integral, or partial derivative. Instead of doing step-by-step it can go to a site (or find a site) and use the calculator, then getting the answer back.

Loguru - Logging framework

A better and more concise way to log in the system. Passing logs to the UI for example.

Flutter based UI

for a native windows app, web-app or mobile app