Skip to platform features
Dotted streams of activity converge through Rysh and emerge as coordinated work
rysh

rysh CLI · the agentic execution environment

The agentic execution environment.
Agents you can see, share and stop.

Every pane is a real shell in its own pseudo-terminal: a person sees and types into it, and a pane can be shared with a teammate who types into it too.

AN OPERATING LAYER FOR AN AI-NATIVE COMPANY

One company.
Many minds in motion.

Bring your channels, specialist agents and people together. Rysh connects the conversation to the work that follows.

01Customer AI — grounded assistants with business tools and human handoff.
02Team & Cloud — live sharing, mirroring and remote access across your team.
03The AI-native company — explore the platform.
Explore the company platform

rysh CLI vs OpenClaw · Hermes · tmux · cmux

Others chat with an agent.
rysh puts it in a terminal you control.

vsOpenClawvsHermesvstmuxvscmux

Four things rysh does in the terminal, side by side with what each product’s own pages say, quoted with the date we read them.

rysh CLIOpenClawHermes Agenttmuxcmux
Panes you share and type into01 · TERMINALReal shells, shareable with a teammate.Every pane is a real shell in its own pseudo-terminal: a person sees and types into it, and a pane can be shared with a teammate who types into it too.A TUI and a browser Control UI. No page we read describes panes or split views.A full TUI. Its agent drives CLIs in a PTY as a tool — not a pane a person sees and types into.Panes, yes. Local users can be given read-only or writable access.Split panes in a native macOS terminal.
Other agent CLIs, governed02 · PROXYClaude Code, Codex CLI, aider — wire-tested.Three vendor agent CLIs — Claude Code, Codex CLI and aider — have been wire-tested through the governance proxy against a logging upstream, and the planted secret appeared zero times in the captured bytes.Its own FAQ sends the coding loop to Claude Code or Codex.Can drive Codex and Claude Code in a PTY (pty=true).Runs them; no agent feature is stated.Runs them “with your own keys”, or through a team model router (CodeRouter).
Policy that fails closed03 · POLICYA missing policy file blocks, it never degrades.Policy fails closed on every agentic path — including a configured org policy file that has gone missing, which blocks execution rather than degrading to no policy.Publishes a policy-as-code design: code-enforced gates, “deny always wins”.Dangerous-command approval (smart / manual / off). No page we read describes policy-as-code.None stated.Enterprise plan lists “shared team rules”.
A human gate before an agent sends04 · CHANNELSNothing posts until a person types send.On Slack, email and WhatsApp, you flip the channel to human and nothing posts until a person types send — enforced in code at the send tool, not asked for in a prompt.Pairing and allowlists for DMs from unknown senders.Draft-and-confirm is a separate plugin for one channel (Telegram Business).No channels stated.Channels and send approval not stated.

Where they lead, said plainly: OpenClaw and Hermes both reach more messaging channels than rysh. Hermes builds skills from experience. Both have far larger communities.

Competitor facts are from each product’s own pages, fetched 2026-10-08 (OpenClaw, Hermes Agent) and 2026-10-07 (tmux, cmux), and re-checked 2026-10-20 for all four.

What you get in every rysh pane

Agents you can see.
Gates you can prove.

01

Shell and agent on one line

In a rysh pane you type shell commands, press Esc twice, and the same line becomes a prompt for the agent working in that directory.

02

Close the terminal, keep the agent

Detach and only the terminal UI closes: the session daemon keeps running, so an agent’s task finishes with no UI attached. Stop a session cleanly and rysh attach restores its tabs, panes and output.

03

Risky calls wait for you

Risky tool calls stop and wait for your approval in the pane: approve, approve always, or reject with a reason. A policy always_gate rule forces the stop even when someone chose approve-always or the run is headless.

04

A fleet, each in its own worktree

Spawn a fleet of agents from files on disk, each in its own git worktree, so they cannot overwrite each other. They report on a shared, threaded board.

05

An agent is a file you can diff

A rysh agent is one markdown file: its persona in the body, its channels in the frontmatter, and its credentials as ${VAR} references resolved at load, so no secret is written into the file.

06

Loops with a written exit

A loop needs a written exit condition or it does not loop, and every loop is bounded on tokens, wall-clock and iteration count — enforced in the execution loop, not requested in a prompt.

07

SecretNAT, on by default

SecretNAT is on by default; you have to turn it off. Secrets are substituted with tokens in the request body before it leaves the machine, and a response carrying a live credential in plaintext is reported into the pane. Responses are not rewritten — by design.

08

Your model, per pane

Bring your own provider — Anthropic, OpenAI, Gemini or a local Ollama — per pane, with streaming. Provider capabilities differ and rysh reports them.

FROM FIRST CONNECTION TO DAILY WORK

CHANNELS → EXECUTION → OUTCOME

Open the front door.
Keep the company moving.

Rysh brings customer channels, specialist agents, connected tools and your people into one operating layer. Conversations carry context into real work; humans can see, approve or take over when needed.

THE CUSTOMER SIDE

One front door.

A branded assistant that knows your business, takes action through connected tools and hands over to a human with the conversation intact.

Website chatEmailMessaging channels
THE TEAM SIDE

One place to act.

People and agents work together in persistent sessions. Share a workspace, connect business APIs and keep consequential decisions with the right person.

PeopleAgentsBusiness systems

RYSH IN MOTION

Less handoff.
More forward.

Follow the work through the runtime.
Every capability, connected.

rysh / executionILLUSTRATIVE WORKFLOW
workspace / workspace

One workspace. Multiple ways in.

01SHELL

Run interactive applications

02PROMPT

Work with an agent

03RYSH

Coordinate the workspace

04CHAT

Talk with your team

01 / A REAL PLACE TO WORK

The terminal.Reimagined.

Real shells. Persistent sessions. A shared workspace where people and agents build side by side.

Native PTYFour input modesLive collaboration
YOUR WORK. YOUR MODELS. YOUR RULES.
A SHARED FOUNDATIONDiscover the platform

From one terminal.
To everything you do.

Build with autonomous agents. Bring a teammate into a live session. Connect your APIs and customers. Rysh gives all of it the same foundation: context, execution and control.

Open CLIYour choice of model

CUSTOMER EXPERIENCE, CONNECTED

ANSWER / ACT / HAND OVER

Not just a better reply.
A useful next step.

Give customers a way to get things done. Give your team the context to help when an automated answer isn’t enough.

01

Answers with a foundation.

Ground your assistant in business content, imported pages and connected knowledge. Give it your tone, language and conversation starters.

02

Tools, not just replies.

Connect order lookup, inventory or your own HTTP tools. Let a conversation reach your systems, with approval where it matters.

03

A person at the right moment.

Review live conversations, guide the assistant privately or take over. Business hours and handoff alerts help route requests to your team.

04

Every opportunity, captured.

Collect leads in the conversation, review them with your team and export them. Use proactive rules to offer help in the right context.

05

See what customers need.

Explore transcripts, conversation volume, unanswered questions and CSAT. Find the gaps in your knowledge and the moments that need attention.

06

Make it feel like your business.

Customize widget appearance, welcome messages and language. Control allowed website origins and manage workspace access and usage.

Meet the conversation where it starts.

Website chat, email, Slack, WhatsApp, Instagram and Discord can connect through Rysh’s chat and channel infrastructure. External channels require adapter configuration and provider credentials.

Explore the connections

SAME FOUNDATION. DIFFERENT WORK.

EXPLORE A WORKFLOW

The work changes.
The context stays.

Customer operations

“Where is my order?”

A customer needs clarity. Your team needs the full picture.

  1. 01

    Understand

    Use the conversation and business knowledge to understand the request.

  2. 02

    Look up

    Call a configured order or shipping tool to retrieve the relevant information.

  3. 03

    Coordinate

    Hand off an exception to a person with the customer’s context attached.

  4. 04

    Resolve

    Reply in the conversation and retain its history for the next interaction.

Illustrative workflows, not live cases. Business actions depend on your connected tools, permissions and team configuration.

THE CAPABILITY ATLAS08 DOMAINS / ONE RUNTIME

A closer look
at what’s possible.

Room for the whole workflow.
Down to the details that make it work.

A terminal made for agentic work.

Keep the fidelity of a real development environment while coordinating multiple streams of work.

+

Tabs, lanes & stacks

Organize parallel work without opening a window maze.

+

Interactive applications

Run vim, htop and full-screen agent CLIs at native terminal speed.

+

Shell · prompt · rysh · chat

Switch the meaning of input per pane without leaving the workspace.

+

Move work live

Rearrange panes while preserving the shell, scrollback and running agent.

BUILT TO CONNECT THE DOTS

A conversation in.
Coordinated work out.

Under the surface, Rysh combines communication, execution and control. The same foundation supports a customer assistant, an engineering workspace or a human-led operational workflow.

Connect the context

Channel adapters, durable events and shared sessions bring messages and work into an addressable runtime.

CHANNELS · NATS / JETSTREAM · SESSIONS

Execute the work

Agents use real terminals, browser actions and API tools. Forge turns your API definitions into reusable capabilities; MCP connects additional tools.

AGENTS · FORGE · MCP · BROWSER

Keep control

Scope access, protect secrets, set usage limits and approve sensitive actions. People can inspect the work and intervene.

POLICY · APPROVALS · USAGE · HISTORY

THE FOUNDATION, TODAY

Start with real capabilities.

Agent execution, real terminals, Forge and browser tools. Website chat, leads, human handoff and analytics. Channel adapters, shared sessions, policies and approvals.

Use the open CLI and shared core, bring your choice of model, and explore self-hosted team deployment.

THE NEXT LAYER

Toward an AI-native company.

Faster company setup, a managed runtime and wider channel coverage are the direction. Phone and IVR have been exercised in a UK local pilot; broader customer rollout remains to be proven.

Product direction and local-pilot status, not a claim of general availability.

CONFIDENCE, BUILT INTO EXECUTION

Let the work move.
Keep the boundaries.

Secrets, tool permissions, approvals and budgets belong in the execution path. Rysh makes those decisions visible as the work happens.

Swap secrets for tokens before a model request leaves your machine.
Approve consequential actions with context.
Track usage and enforce prompt budgets.
Inspect history and the audit trail.

BEFORE YOU BEGIN

A little
more context.

Read the documentation

Rysh connects conversations to execution. Customer assistants, human teams and AI agents share a foundation of context, tools and control. Its real-terminal runtime supports software work, connected business processes and human-led operations — with a broader vision of an AI-native company.

YOUR NEXT CHAPTER

Put your agents
to work, together.

Private build (ry)

brew install rysh-ai/rysh/ry

Local execution. Your choice of model.