Agent Sessions that pick up where you left off.
Run managed agents in a persistent workspace. Configure a provider for your project, choose an agent, and continue the same conversation across Turns. Hosted environments run on run9; you can also connect a local machine as an Env.
One workspace. Conversations that continue.
Create an Env, then create a Session for an agent in that workspace. Each input runs a Turn; later Turns resume that Session's conversation. Idle Sessions do not need a live agent process between Turns.
Install the sys9 CLI and sign in. Before running the example,
use project settings to configure and activate
a Codex provider. provider ls shows the current configuration; the Env
and Session commands return the IDs used in later steps.
# sign in, then configure a Codex provider in project settings $ sys9 smith auth login $ sys9 smith provider ls # create a shared workspace for your Sessions $ sys9 smith env create --name workspace # replace ENV_ID and SESSION_ID with the returned IDs $ sys9 smith session create --env ENV_ID --runtime codex --name first-task $ sys9 smith session run SESSION_ID --input "Describe the workspace" # continue the same conversation $ sys9 smith session run SESSION_ID --input "Suggest a next step"
Env → Session → Turn.
A workspace holds the files. A Session holds the conversation. A Turn does the next piece of work.
The workspace
An Env provides a persistent workspace and runtime home shared by its Sessions. Hosted Envs use managed runtimes; a connected local Env uses its own installed agents and provider configuration.
The conversation
A Session belongs to an Env and preserves an agent conversation across Turns. Create separate Sessions for separate conversations while using the same Env files.
The next execution
A Turn runs one input in a Session. Successful Turns advance its saved conversation; failed or stopped Turns do not advance continuation. A new input activates the runtime when needed.
Choose the runtime for the task.
The managed catalog includes Codex, Claude Code, Gemini CLI, Pi, and OpenCode. Configure the provider for each agent you use; available models depend on that provider.
Build around your agent service.
Use each service's own API and access configuration where it fits your application.
Linux environments
The compute substrate behind hosted smith9 Envs.
run9 → task9Shared tasks
Add a task queue to an application that coordinates agent work.
task9 → inbox9Messaging
Add agent mailboxes through the inbox9 API.
Realtime → chord9Agent workbench
Collaborate on native coding agent Sessions across local and cloud Projects.
chord9 →Start a Session on smith9.
Sign in, select your project, and configure an agent provider. Create an Env and send the first input through the portal or CLI.