Skip to main content
r5dchat is a compiled binary at /usr/local/bin/r5dchat in the workspace image. It runs the workspace daemon and gives the agent read access to its own conversation history.
convo is an alias for conversation.

​
r5dchat daemon

Starts the workspace daemon. This is the container’s default command, so you rarely run it manually except in local development. The daemon listens on port 7337 and serves an authenticated private HTTP protocol for file, search, and process operations, plus an unauthenticated /healthz endpoint used by the readiness and liveness probes.
​
string
default:"/home/r5d"
Root directory the daemon operates in.
​
integer
default:"7337"
Port the daemon listens on.
​
string
required
Bearer token required on every request. Supplied from the workspace identity Secret in Kubernetes.
​
string
Workspace identity presented when calling back into the control plane, which is how the conversation commands authenticate.
​
string
Control-plane base URL used by the conversation commands.
​
string
The owning user’s ID, set by the controller.
Run it locally with:

​
Conversation commands

These let the agent recover context after a fresh-context boundary without keeping the entire transcript in the model’s context window. Requests authenticate with workspace identity, and the server independently re-checks that the workspace’s user owns the requested conversation.

​
overview

Returns a condensed map of the conversation: its turns, their IDs, and short summaries. This is the recommended starting point after a fresh context, because it is small enough to read cheaply.

​
turn

Returns the full detail of one turn, including tool calls and results. Use it to drill into a specific turn identified from the overview.

​
read

Returns the complete transcript in Markdown, with fresh-context boundaries marked. Use it as a fallback when the overview is not enough, keeping in mind that a long conversation can consume a large amount of context.
r5dchat conversation <conversation-id> without a subcommand still performs a full read, for backwards compatibility.

​
Output format

Human-readable Markdown by default:
Pass --json for machine-readable output when scripting.

​
Typical agent workflow

1

A fresh context begins

The flagged system prompt instructs the agent to orient itself before continuing.
2

The agent reads the overview

bash r5dchat conversation overview abc123
3

It drills into relevant turns

bash r5dchat conversation turn abc123 turn-7
4

It falls back to a full read only if needed

bash r5dchat conversation read abc123
This keeps the durable transcript complete while the provider context stays small.