Frequently asked questions
The short version: your agents, your machine, your keys.
Where does my code run?+
On your own machine. The jumpterm daemon runs the agent locally on your files; the web and iOS apps just relay to it. Your code never lives on our servers.
Do my API keys leave my machine?+
No. BYOK keys are encrypted at rest and injected into the agent on the machine that runs it. They're never sent to a browser or phone in plaintext — when you launch from the web, the key goes straight to your daemon, not back to you.
How do you know a key works before I run a session?+
When you add a key we verify it live against the provider and show Connected ✓ (or a clear 'invalid key' error). The model picker also tells you when an agent needs a key before you launch.
Which agents and models are supported?+
Claude Code, Codex, OpenCode (which routes to 75+ providers), and a custom shell adapter. See the full list on the Models page. Free local models work via Ollama.
What happens if my machine goes offline mid-session?+
The session pauses (it's not lost — it's event-sourced). The apps show that the host went offline, and you can re-launch onto any online host when you're back, resuming where the agent left off.
Can I see how much I'm spending?+
Yes. Each session shows its token use and estimated cost, and Settings has a per-agent usage rollup over 7/30/90 days. We use the provider's reported cost when available, otherwise estimate from list prices.
What is Jump OS?+
Jump OS is the local optimizer your sessions run through, on your own machine. It meters real usage, caches and dedupes the calls your agents make, maximizes provider prompt caching, and can enforce a per-session token budget — so you get more work out of the same plan. Your key is held only in memory by Jump OS and never enters the agent's environment. Toggle it per session in the composer; see what it saved in the session view.
Will Jump OS keep me from getting rate-limited?+
It helps two ways. First, its caching and dedup reduce the tokens you actually spend, so you burn your plan's rolling limit slower. Second, Settings → Usage window tracks your live burn rate and — if you tell it your plan's token ceiling — forecasts when you'll hit it ("~24m to your limit at this rate"), so you can pause or pace heavy sessions before you're cut off. We don't have a provider API for your exact remaining quota, so the forecast is based on your own measured usage, not a guess.
Does Jump OS send my data anywhere?+
No. Jump OS runs inside the daemon on your machine. It sits between your agent and the provider you already use, forwarding to that provider with your key — nothing is routed through us, and your code and tokens never leave the box.
Can I self-host the relay?+
Yes — point the CLI at your own relay with
JUMPTERM_API_URL. See the docs.How much does it cost?+
Jumpterm is free during beta — you only pay your LLM provider for what you use. See pricing.
Can I control what an agent is allowed to do?+
Yes. Pick a permission mode per session (from read-only planning to auto-accept edits), and approve or deny individual commands and file edits from any device — with the real diff in front of you.