AI Assistant

How JARVIS reads a request, follows the conversation, and uses its capabilities when a reply needs them.

JARVIS is a conversational assistant. You write in plain language, it works out what you are asking for, and it answers in the context of everything that came before.

What JARVIS can do

Answer questions across a wide range of topics, and adapt the depth to what you asked for.
Follow the conversation. It reads the thread, not just your latest message.
Handle follow-ups that only make sense in context.
Work through complex tasks — reasoning, comparing options, reviewing code, drafting, summarising.
Use its capabilities when a reply needs them, rather than answering from memory alone. See Web Search, Memory, and Tasks.
Reply in your language. Write in German and it answers in German.

Following context

JARVIS carries the thread forward, so follow-ups can be short and referential.

You: Tell me about the main differences between REST and GraphQL.
JARVIS: explains

You: Which one would you pick for a mobile-first app?
JARVIS: recommends, and says why — using the comparison it just made

You: And what about caching?
JARVIS: continues from the same discussion

You do not need to restate the subject. Phrases like "that", "this", "the other one", "what about it?", and "and now?" resolve against the conversation.

Using capabilities when appropriate

JARVIS decides per message whether a capability is needed.

Message	Behaviour
"Explain database indexes"	Answered directly
"What's the latest Python version?"	Searches first, answers with sources
"Remember that I prefer dark mode"	Saves a memory
"Add a task to renew the domain"	Adds to your task list
"Forget that I prefer dark mode"	Removes that memory

You can always be explicit:

Search this on Google: Node.js release schedule.
Add a task: read the OpenAPI spec.
Remember that our API is versioned monthly.
Forget my old project notes.

What happens when something isn't available

JARVIS distinguishes between I can't do that and that isn't available right now:

Outside a capability — a direct answer that it isn't connected yet, no error, no pretending.
A tool that failed — a brief note that it couldn't complete the lookup, then an answer from what it knows, flagged as possibly out of date.
A policy denial — one fixed notice. See JARVIS Policies.

JARVIS will not invent citations, fabricate a search it did not run, or present a guess as a verified fact. When it is uncertain, it says so.

Never in a reply

JARVIS does not disclose credentials, keys, session tokens, or internal system details — not when asked directly, and not through indirect phrasing. It also cannot change your account by itself: if it detects a preferred-name statement, it offers a confirmation prompt and waits for you.

Safety

Requests that would help cause harm are declined with a short notice. Ordinary questions, legitimate research, and defensive discussion are all welcome — including about security topics. Full guidelines are on the JARVIS Policies page.

Conversations — how threads are saved and resumed
Web Search — when and how searching happens
Memory — what persists between conversations
Tasks — driving your task list from chat