Comparison

Tryssh vs Warp

Warp is the closest thing to us in spirit: a terminal with a real agent in it, on three platforms, with a permission system its authors clearly thought about. The difference is what the app is organised around — a shell session, or a host.

Warp facts verified as of July 29, 2026 · Tryssh 0.1.6 · macOS 14+

What each one is for

Warp is a terminal first. It documents support for macOS (Intel and Apple silicon) 10.14 or above, Windows 10 version 1903 and later plus Windows Server 2022, and Linux on x86_64 and ARM64 with glibc 2.31 or newer. Inside it sits Warp Agent, with conversations tied to sessions and multiple Agent Mode conversations running simultaneously across windows, tabs and panes. Your working unit is a shell session, wherever that shell happens to be.

Tryssh is an SSH workspace. Your working unit is a host. Each machine in the list gets its own terminal, its own chat with its own memory, its own live metrics, its own background jobs and recurring checks, and its own audit trail of every command the agent proposed and what you did about it. The agent is per-host because the questions are per-host: "why is this box at 89% disk" is not a question about a shell session.

If you spend your day in a terminal writing code, Warp is aimed at you. If you spend your day on twenty machines that are not your laptop, we are.

Verified as of July 29, 2026. Every Warp claim below is taken from warp.dev or docs.warp.dev, with the exact source URL in an HTML comment next to the claim in this page's source. Warp ships fast — if you spot something stale, tell us.

Capabilities

Side by side

Read the "Warp" column first. It is longer than ours on purpose.

Capability Warp Tryssh
Platforms Yes macOS 10.14+, Windows 10 1903+ and Server 2022, Linux (glibc 2.31+) No macOS 14+ only (Apple silicon, arm64)
Shape of the app Terminal with blocks, panes and tabs; agent conversations are tied to sessions Host list; each host has a terminal, a chat, metrics, files, jobs and watchers
Saved SSH host manager Not documented — you run ssh as you would in any terminal Yes First-class: host records in SQLite, secrets in the macOS keychain
Anything installed on the remote host Yes for the SSH extension: a writable home for a ~/.warp*/ install, fetched over HTTPS from app.warp.dev or uploaded over SSH Yes Nothing. Plain SSH — an ssh2 exec channel and a PTY
Remote host requirements For the SSH extension: macOS or Linux, x86_64 or aarch64, bash or zsh, glibc 2.31+ (so no CentOS 7/8) An SSH server. Old boxes, BSDs and appliances included
Remote code editing & codebase indexing Yes Remote file tree, editor, diffs, and semantic search over the remote repo Partial: SFTP browser, a remote file editor, and agent read/write/patch on single files. No indexing, no semantic search
Agent auto-run default Command allowlist is empty by default; autonomy levels include "Agent decides", where the agent acts autonomously when confident Default-deny by pattern: auto-runs only a read-only allowlist with no shell metacharacters. No confidence-based mode
Always-prompt denylist Yes Denylist takes precedence over both the allowlist and "Agent decides" Yes Denylist always prompts, even with auto-approve on for that host
Permission granularity Yes Per-action autonomy, reusable agent profiles, separate MCP allow/denylists Coarser: per-host autonomy plus a per-host command allowlist you build as you approve
Model choice Yes Models from OpenAI, Anthropic and others No Anthropic only
Free plan $0 with limited cloud agent access; telemetry must be enabled to use AI features on Free $0 with 200 agent credits to start — one-time, nothing refills it. Terminals, SSH and SFTP are never metered
Paid entry price Build from $20/month for 1,500 credits; Max from $200/month for 18,000; Business from $50/user/month Pro $19/month for 2,400 agent turns; Scale $29/month for 4,500. Not per seat
Teams Yes Business plan with team usage metrics and SAML SSO; Enterprise for larger orgs No Single-operator today. No sharing, no sync, no admin panel
Where host records live No host records to store — SSH config stays wherever your shell keeps it SQLite under ~/.tryssh; keys and passphrases in the macOS keychain, never in a database
Command audit trail Not documented as a queryable log Yes Every agent command with its risk level, whether it was auto-approved, exit code and output — local SQLite
Bring your own model key Not offered — agent usage is sold as plan credits Yes With ANTHROPIC_API_KEY in the core's environment, chat talks to Anthropic directly — no gateway, no credits, no account
Maturity Yes Years of shipping, three platforms, a large user base No 0.1.6, first shipped 2026

Where Warp is stronger

This is the comparison page where we lose the most rows, and we would rather write them down than let you find them.

  • Platform coverage. macOS, Windows and Linux. We are macOS 14+ and nothing else. If half your team is on Windows, stop reading.
  • Maturity and polish. Warp has been shipping a GPU-accelerated terminal for years across three platforms. Tryssh is 0.1.6. Maturity is a feature and we do not have it yet.
  • The permission system is richer than ours. Per-action autonomy levels, reusable agent profiles, separate MCP allow and denylists, and team-level rules an admin can enforce. Ours is one autonomy setting per host plus an allowlist you grow by approving things. Warp's is better engineering; ours is a stricter default.
  • Remote code editing and codebase indexing. Warp's SSH extension gives you a remote file tree, an editor, native diffs, and semantic search over the remote repository. We have SFTP and single-file edits. That is not the same tool.
  • Model choice. Warp offers models from OpenAI, Anthropic and others. Our agent is Anthropic only.
  • Teams. A Business plan with team usage metrics and SAML SSO, and an Enterprise tier above it. We have none of that.

Note also that Warp's Business plan is priced per seat and ours is not — and that Build starts at $20/month for 1,500 credits against our 2,400 at $19. A Warp credit and a Tryssh turn are not the same unit, so treat that as a rough shape, not a benchmark.

Where Tryssh is different

1. The host is the object, not the session

Warp organises around shell sessions; agent conversations are tied to sessions. Tryssh organises around machines. A host has a saved record, a stored credential reference, a persistent chat with memory of what you did on that box last week, sampled CPU/memory/disk/network metrics, background jobs that survive the command taking twenty minutes, and recurring watchers the agent can set up for a check you want repeated.

The practical consequence is what happens when you come back on Thursday. In a session-shaped tool, Tuesday's context is gone with the tab. In a host-shaped tool, you open the host and the conversation is still about that host.

2. Nothing goes on the remote machine

Warp's richer SSH features are documented as requiring a supported remote: macOS or Linux, x86_64 or aarch64, bash or zsh, glibc 2.31 or newer, a writable home directory for a ~/.warp*/ install, and, to install its remote agent, either outbound HTTPS to app.warp.dev or an upload over SSH; core features then work without outbound access from the host. That buys real capability and it is a fair trade for a laptop-like dev box.

It is a worse trade for a locked-down production host, an appliance, a CentOS 7 box someone will not let you touch, or anywhere a binary in ~/.warp*/ is not allowed to sit. Tryssh installs nothing and requires nothing but an SSH server: the agent uses an ssh2 exec channel, and your interactive shell is a separate PTY. Those two channels are architecturally distinct — the agent cannot read your terminal or type into it.

3. Default-deny, without configuring anything

Warp's command allowlist is empty by default, and its autonomy options include "Agent decides", where the agent acts autonomously when it is confident and prompts when it is uncertain. That is a reasonable design, and if you configure it carefully it is stricter than most tools ship.

Tryssh has no confidence-based path. A command auto-runs only if it matches a read-only allowlist — ls, cat, df, ps, journalctl, systemctl status, git log, kubectl get and friends — and contains no shell metacharacters. Anything unrecognised is treated as unsafe, including a command whose quoting does not balance. Everything else renders the exact command on a card and waits for your click. A separate denylist — mkfs, dd of=, recursive rm against a system path, flushing firewall rules, stopping the sshd you are connected through, piping a download into a shell — always asks, even if you turned auto-approve on for that host.

Then every one of those commands lands in a local audit log with its risk level, whether it auto-ran, its exit code and its output.

Not a claim about safety in general. A default-deny gate is a real constraint, not a guarantee. It stops an agent surprising you; it does not stop you approving something you should not have. You are still the operator.

4. Local-first data, with one honest asterisk

Hosts, chats, metrics and the audit log live in SQLite under ~/.tryssh. SSH keys and passphrases live in the macOS keychain under the service ai.definable.tryssh and never touch a database. SSH traffic goes from your Mac to your servers; we are not in that path.

The asterisk: when you use the hosted copilot, that turn's conversation — including the command text and the command output the agent read — is relayed through our gateway to Anthropic to generate the reply. That is how the agent works, we say so in the privacy policy, and we do not train on it. Run the core with your own ANTHROPIC_API_KEY and the gateway is out of the picture entirely. Warp, for its part, documents that with telemetry disabled no console interactions are persisted on its servers, and offers a Zero Data Retention agreement on its Business and Enterprise plans.

Migration

Trying Tryssh alongside Warp

"Alongside" is deliberate. These tools are not mutually exclusive and we are not asking you to uninstall anything.

Tryssh has no importer. There is no ~/.ssh/config import and no import from any other tool. Hosts are added by hand in the app. If you keep fifty entries in your SSH config, budget for the typing — or add five and judge from those.

  1. Keep Warp as your terminal

    It is a better terminal than we are, on more platforms, and nothing about installing Tryssh interferes with it. Use us for the machines, not for local development.

  2. Add the three or four servers you actually babysit

    Not the whole fleet. The ones you SSH into when something is wrong. That is where a per-host agent with per-host memory pays for itself.

  3. Point at a key rather than pasting one

    Add a private key once to the key vault and reuse it across hosts; the key bytes and any passphrase go to the macOS keychain and the database stores only a reference. ssh-agent also works — pick "SSH agent" as the auth type and Tryssh uses your existing SSH_AUTH_SOCK.

  4. Give it a real incident, not a demo

    "Why did nginx start 502ing at 4am" tells you more in three minutes than an hour of reading comparison tables, including this one. The commands it proposes are the product.

FAQ

Questions, answered

Is Warp an SSH client?
Warp is a terminal, and ssh works in it as it does in any terminal. Its richer SSH features — remote file tree, remote editing, codebase indexing — are documented as needing a supported remote host: macOS or Linux, x86_64 or aarch64, bash or zsh, glibc 2.31+ on Linux, a writable home for a ~/.warp*/ install, and, to install its remote agent, either outbound HTTPS to app.warp.dev or an upload over SSH; core features then work without outbound access from the host. Tryssh installs nothing on the remote and works over plain SSH.
Does Tryssh run on Windows or Linux?
No. macOS 14+ only, Apple silicon (arm64) — there is no Intel build. Warp documents macOS 10.14+, Windows 10 1903+ and Server 2022, and Linux with glibc 2.31 or newer. Windows and Linux builds of Tryssh are planned, not shipped — do not buy on that promise.
How is your approval gate different from Warp's?
Warp's command allowlist is empty by default and its autonomy options include "Agent decides", where the agent acts autonomously when confident. Tryssh has no confidence-based mode: auto-run requires a read-only pattern match and no shell metacharacters, and anything unrecognised is unsafe. Warp's system is more configurable and enforceable across a team; ours is stricter with zero configuration.
Can Tryssh import my hosts?
Not today. No ~/.ssh/config import, no import from any other tool. Hosts are added by hand. It is the biggest hole in our migration story and it is on the list.
Can I use both?
Yes, and most people evaluating us should. They do different jobs: Warp is where you write code, Tryssh is where you look after servers. Neither one touches the other's configuration.
What actually leaves my machine?
SSH traffic never does — it goes straight from your Mac to your servers. Keys and passphrases never do; they stay in the macOS keychain. When you use the hosted copilot, that turn's conversation, including command text and command output, is relayed through our gateway to Anthropic to generate the reply; we do not train on it. Details in the privacy policy. Run with your own ANTHROPIC_API_KEY and the gateway is skipped entirely.

Add one server. Ask one real question.

That is the whole evaluation. 200 agent turns to start, no card required.