coming soon

private chat.
verifiable.

your chats stay on your device. your queries run inside tamper-proof hardware. every response is cryptographically signed — and verified on your iphone before you read it.

xchacha20-poly1305·intel tdx·nvidia gpu tee·swiftdata local
wire view
wire view
what you need to get started

teemoon is a client. you bring the model.

the app is free. you fund providers directly. nothing in between.

· required01
a near.ai api key
near.ai is the recommended starting point — it's the only provider that runs models inside a tee, which is what makes teemoon's privacy guarantees verifiable end-to-end. create an account, fund it, generate a key. any other openai-compatible endpoint works too if you already have one.
near.ai · recommended · ~2 min signup
· recommended02
a brave search api key
optional, but highly recommended — it grounds responses in live web results with citations. brave's llm-grounding endpoints are the only ones that pair cleanly with teemoon's privacy model. takes about a minute to set up.
brave search api · free tier available
· then03
paste, verify, chat
open teemoon, paste your keys into the keychain-backed setup screen, and start chatting. your keys never leave your device. the app talks directly to each provider — no teemoon relay.
~ 2 minutes end-to-end
what the wire sees

this is what you type. this is what the wire sees.

both panes are from the live app. the left is what you type. the right is the same request, captured by the in-app debug panel, exactly as it leaves your iphone — all user content sealed before transit.

header
X-Client-Pub-Key
ephemeral key generated on your iphone for this session.
header
X-Model-Pub-Key
the model tee's key, verified via intel tdx attestation.
header
X-Encrypt-All-Fields: true
client and model derive a shared secret via ecdh. the gateway routes ciphertext but can't read it.
body
every content, name & arguments is a sealed blob
structural metadata (role, model) stays plaintext — it contains no user data.
tool
even tool names are encrypted
the gateway can't see what tools you're calling.
header
X-Client-Pub-Key
ephemeral key per session.
header
X-Model-Pub-Key
the tee's key, verified via tdx attestation.
header
X-Encrypt-All-Fields: true
client+model derive a shared secret. gateway routes ciphertext.
body
every content is a sealed blob
only role and model stay plaintext — no user data.
tool
even tool names are encrypted
the gateway can't see what tools you're calling.
settings → developer → debug panel · open it any time to verify.
security

up to seven cryptographic checks. every reply.

before any response reaches you, teemoon verifies the hardware it ran on, the chain of trust above it, and the signatures on the bytes themselves. the math runs on your iphone — not on a server you have to trust.

all 7 checks passed
verified just now · 142 ms on device
gateway hardware
tdx quote signature
intel tdx confidential compute
ok
intel certificate chain
root → platform → quote, full chain
ok
signing key bound to hardware
gateway key derived inside enclave
ok
encryption
end-to-end encryption
xchacha20-poly1305, 256-bit keys
ok
model tee verified
nvidia gpu confidential compute
ok
model key proven on-device
session keys provable on-device
ok
runtime
response signatures
every chunk signed, verified inline
ok
brave grounding

the model does the searching.

most chat apps do one search at the start of the turn and call it done. teemoon hands the search tool to the model itself — it writes its own queries, fires several in parallel, reads the results, and searches again if it needs more. you see every query and every source as it happens.

model writes the queriesmulti-step lookupsparallel fan-outinline citations
model is searching·3 queries · 12 sources2.4s
01
"shimokitazawa indie show calendar may 2026"
fresh:7d · 0.4s
02
"shelter shimokitazawa lineup may 6 7 9"
fresh:7d · 0.6s
03
"basement bar tokyo schedule may"
fresh:7d · 0.5s
→ model decided one search wasn't enough. fired two follow-ups based on what it found. cited 12 sources inline.
every query is yours to inspect
providers

five providers. or any openai-compatible endpoint.

switch models on a tap. bring your own key with zero markup. your device talks directly to each provider — teemoon never sees the traffic.

tee
near.ai

hardware-attested ai with end-to-end encryption. cryptography, not promises.

only tee-verified provider
new
grok / xai

xai's truth-seeking models. grok 4.3 at lower prices, up to 2 million token context.

requires xai api key
fast
fireworks

open-weights at speed. kimi, deepseek, glm, qwen, llama, and more.

requires fireworks api key
web
brave answers

llm answers grounded in the live web with inline citations.

requires brave api key
custom
byok

any openai-compatible endpoint. keys live in your keychain.

requires your own key
models

the best models. routed privately.

new models drop weekly. tee-attested when available. all of them tap-switchable inside one chat.

bring your own key. any openai-compatible endpoint works — plug in your preferred provider and model.

modelorgsizeviatee
glm 5.1z.ai754bnear.aie2ee
qwen3.5alibaba122bnear.aie2ee
grok 4.3xai500bxai—
kimi k2.6moonshot1tfireworks—
deepseek v4 prodeepseek1.6tfireworks—
qwen3.6 maxalibaba1t+ moefireworks—
minimax m2.7minimax230b moefireworks—
near.ai
glm 5.1
754b · z.ai
e2ee
qwen3.5
122b · alibaba
e2ee
xai
grok 4.3
500b · xai
fireworks
kimi k2.6
1t · moonshot
deepseek v4 pro
1.6t · deepseek
qwen3.6 max
1t+ moe · alibaba
minimax m2.7
230b moe · minimax
zero data collection

we built it so we couldn't read it, even if we wanted to.

no accounts. no telemetry. no cloud sync. chats live in swiftdata on your iphone — full stop. the gateway sees ciphertext; the model runs in a tee. keys never leave the tee.

0
accounts created
0
analytics events
0
bytes stored on our servers
∞
chats kept on your device
faq

questions, answered.

two things. (1) an api key from any openai-compatible provider — we recommend near.ai because it is the only one running models inside a tee, which is what makes the privacy guarantees verifiable. you create an account, fund it, and generate a key. (2) optionally, a brave search api key for grounded web answers — the response quality is dramatically better with it. setup takes about two minutes total. the app is free; you fund providers directly.