Asks first
Manual waits for your yes on every edit and command. Plan mode reads and proposes; nothing is edited until you approve the plan.
Bring your own agent
Runs the agents you already use.
termique finds the agent CLIs on this computer or on the host and runs them with your own login, plan and limits. Chat with the ones that have a protocol for it, or open any of them in its own terminal UI. If one is missing, it shows you the install command.
Found on this computer
Claude CodeAnthropic · chat or its terminal, agent teams too
InstalledCodexOpenAI · chat or its terminal
InstalledGrokxAI · chat · experimental
InstalledGemini CLIGoogle · chat or its terminal · experimental
InstalledOpenCodeOpen source · chat or its terminal · experimental
InstalledAntigravityGoogle · its terminal · experimental
InstalledKilo CodeOpen source · its terminal · experimental
InstalledHermesNous Research · its terminal · experimental
Install$ curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
All sessions
Every agent, on one board.
Sessions from every project and host, sorted by what they need from you. Idle ones pause to save memory and pick up where they left off.
All sessions · 63 projects · 2 machines
Needs you 2
Fix token refresh
Wants to run a command:git commit -m "fix(auth): …"Allow onceDeny
Fix flaky checkout test
Should the fake clock also cover the discount timer, or only the cart?Reply
Working 2
Release v0.8.2
Running npm testRate limiting for /login
2 subagents exploring the auth flowDone 1
Copy on select
+179 −34 · 4 filesvitest 572 passed · tsc okReview changes
Idle 1
Upgrade Postgres client
Paused to save memoryYou stay in charge
Nothing runs until you say yes.
New sessions start in Manual: every edit and command waits for you. Loosen it per project, one rule at a time.
Permission request~/Codes/termique
Claude Code wants to run a command
git commit -m "fix(auth): refresh the token before it expires"Allow onceAlways allow git commitDeny
Or tell it what to do instead.Default modetermique
ManualPlanAutoFull access
Always allowedin any mode
git status
git diff *
npx vitest run *
cargo test *
Its own worktree
Each session works on its own agent/ branch in its own git worktree, so two agents never edit the same checkout.
Review, then commit
Read the diff, leave notes on lines and send them back. A verify gate runs your tests and typecheck before a session can call itself done.
Knowledge
Agents that know the project before you type.
AGENTS.md and pinned docs go into every session; the rest is searched when a task needs it. After each session termique suggests what it learned, and you approve each one.
Docs & memorytermique
AGENTS.mdpinned · every session
Release runbooksearched · used 3×
Design system notessearched · used 9×
How I like to workyours · every project
Learning to saveThe authStore timestamp test is flaky. Re-run once before trying to fix it.
SaveNot now
Skills built into termiquemade for servers
server-triageFinds why a host is slow or unhealthy with read-only checks.
deploy-checkHealth endpoints, error rate and disk, before and after a deploy.
log-investigateSearches journald, nginx and app logs around an incident.
nginx-safe-editEdits config, runs nginx -t, reloads only if the test passes.
docker-opsInspects containers and restarts one service at a time.
migration-safetyFlags long locks, table rewrites and missing down steps.
ssh-hardeningAudits sshd_config, keys and fail2ban without locking you out.
incident-reportTurns an investigation into a postmortem.
Works with Claude Code, Codex and Gemini CLI. Add your own, or import a folder of skills.
On your servers
Runs where your code runs.
A project can be a folder on a host. termique connects over SSH, the same way it already does, and keeps the agent running there.
Close the laptop. It keeps going.
Remote projects run inside tmux on the host, so a dropped connection or a sleeping laptop doesn't stop them. No tmux? termique offers the install command, or runs without it for a short task you'll watch.
Agents on a schedule. Experimental.
Run a prompt on a schedule, with a cheap check first so it only runs when there's something to do. On a host it runs even while termique is closed, and each run is a session you can open and continue.
Morning log triage
Only run if: journalctl -u api --since yesterday -p err -q | grep -q .One slow query on /orders, 1.8 stoday 09:00
Skipped: no errors overnighttue 09:00
From your phone
Approve from the queue at the coffee shop.
Sign in on the termique app and your sessions are there, live. Answer a question, allow a command, or start a new session while you're away.
View only until you allow itA new sign-in sees your sessions but can't run anything until you approve that phone on your computer.
Face ID for risky commandsAnything that deletes, pushes or runs on a host asks for Face ID first.
End-to-end encryptedStatus, chat and terminal are decrypted only on your devices. termique's servers pass messages on but can't read them.
9:41Agents
Needs you · 2WorkingAll
Fix token refresh
git commit -m "fix(auth): …"Allow onceDeny
Fix flaky checkout test
Should the fake clock also cover the discount timer?Release v0.8.2
Race · experimental
Not sure which agent? Race them.
Send one prompt to several agents, each in its own worktree. termique runs the same checks on every result and recommends the one to keep, or mixes the best parts on a new branch.
Fix flaky checkout test3 worktrees from main
| Check | Claude Code | Codex | Gemini CLI |
|---|---|---|---|
| Verify gate | 48/48tests · typecheck ok | 48/48tests · typecheck ok | 46/482 tests fail |
| Flaky check | 20/20passed | 19/20passed | —skipped, gate failed |
| Change size | +42 −83 files | +118 −316 files | +27 −42 files |
| Second opinion | 0 gaps | 1 gapchanges cart's public API | 2 gapshides the race with a delay |
| Time | 6m 12s | 8m 40s | 4m 05s |
Recommended: Claude Code. Passes every check, stable across 20 runs, the smallest change with no gaps.Mix the best partsKeep this one
Supervisor · experimental
Someone watching while you're not.
A supervisor keeps an eye on every session: it approves the routine by your rules, breaks loops, keeps context small and hands work to another agent when a limit runs out.
Today · watching 4 sessions23 approvals saved
Paused a loop
Codex ran npx vitest run authStore three times and got the same failure. It was about to try a fourth fix to the same line.
Handed off to Codex
Claude Code reached 95% of its 5-hour limit. Codex picked up from a written brief on the same branch, and hands back after the reset.
Compacted context
Context hit 86%. Kept your requests, the plan and the changed files; dropped 31 old command outputs. Now at 38%.
Approved 23 routine requests
Tests and diffs in termique, under your rules. Nothing on a host.
Asked you instead
rm -rf node_modules dist matches no rule and deletes files, so it went to you.
Your rulesper project
Same error repeatspause and ask · 3×
No progress15 min
Compact contextat 85%
Hand off to another agentat 95%
Push to your phoneafter 10 min waiting
FAQ
Questions, answered.
What Agents is, where it runs and what it can touch.
01Does termique come with its own AI model?
No. It runs the agent CLIs you install, with your own accounts and limits: Claude Code, Codex, Grok, Gemini CLI and OpenCode in a chat, and Antigravity, Kilo Code, Hermes and Claude agent teams in their own terminal.
02Where do the agents run?
In a project folder on your computer, or in a folder on a host over SSH. On a host they run inside tmux, so they keep going if you disconnect.
03Can an agent change things without asking?
Not by default. New sessions start in Manual mode, where every edit and command waits for you. You decide per project what's always allowed.
04Can termique read my code or my chats?
Sessions run on your own machines. When they're mirrored to your phone, messages are end-to-end encrypted; termique's servers relay them but can't read them.
05What does it cost?
Agents comes with termique. Free runs one session at a time on your computer. Pro, the same $5 a month, runs sessions side by side and on your hosts, races agents on one task, and adds the Supervisor and scheduled automations. Following and approving from your phone works on both.
06What's experimental in the beta?
Races, the Supervisor and scheduled automations, which are off until you turn them on in Settings › Agents. And every agent besides Claude Code and Codex: Grok, Gemini CLI, OpenCode, Antigravity, Kilo Code, Hermes and Claude agent teams. They work, but get less testing during the beta.
07Can I send it a screenshot?
Yes. Paste one, drop images from Finder or attach them, up to ten per message. Claude Code, Codex, Grok, Gemini CLI and OpenCode get the picture itself, on a host too.
08When can I use it?
Agents is in private beta. Join the waitlist and we'll send you one email when your spot is ready.