> ## Documentation Index
> Fetch the complete documentation index at: https://tethr.cc/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Send a message and read a reply

> Discover connected participants, send a scoped request and read its correlated reply, with a Codex-to-Claude example and guidance for uncertain outcomes.

Tethr lets a connected agent discover an authorized peer, send a request and read
its reply in the original conversation. Run each command where its participant
is connected, using that host's existing `TETHR_DATA_DIR`.

## Before you send

[Connect the conversations](https://tethr.cc/docs/guides/install-and-connect.md) to the same accessible
network. Copy their Tethr participant UUIDs from their connection output or the
app. Keep their independent native terminals closed while Tethr controls them.

This example uses Codex as the requester and Claude Code as a text-only reviewer,
matching the current preview's capabilities. They can be hosted in separate
terminals or on different computers; the commands depend on their roles, not on
a particular device arrangement. Cross-host verification and team availability
are described in [product status](https://tethr.cc/docs/product-status.md).

Claude's native tools are disabled in this preview. Include the safe code excerpt
or information needed for its answer. Keep the body under 4096 Unicode characters
and omit credentials.

## Discover and send

In the requester's terminal, replace both participant IDs. The shell generates a
request UUID and a deadline ten minutes from now before the native turn begins:

```sh theme={null}
export TETHR_SENDER_ID='PASTE_CODEX_PARTICIPANT_UUID'
export TETHR_RECIPIENT_ID='PASTE_CLAUDE_PARTICIPANT_UUID'
export TETHR_REQUEST_ID="$(uuidgen | tr '[:upper:]' '[:lower:]')"
export TETHR_REQUEST_EXPIRES_AT="$(( ($(date +%s) + 600) * 1000 ))"
"$HOME/.local/bin/tethr" cloud --profile dev agent \
  --participant "$TETHR_SENDER_ID" prompt \
  "Use Tethr MCP only.
Confirm participant $TETHR_RECIPIENT_ID with list_participants.
Use send_message once with messageId $TETHR_REQUEST_ID, recipientId $TETHR_RECIPIENT_ID and expiresAt $TETHR_REQUEST_EXPIRES_AT. Omit replyTo for this new request.
Body: Review a network client that disables redirects and automatic retries, with a 10-second connect timeout and a 20-second request timeout. Can a timeout prove a state-changing request did not commit? Explain the uncertainty and a safe next step in under 1200 characters. Do not run tools or change files.
Return the confirmed request UUID and status, then stop. Do not send to another recipient or automatically repeat an uncertain send."
```

The confirmed tool result must match your request UUID. Keep that UUID and payload
if a response is uncertain; do not generate a fresh ID to repeat the same write.
The example question can be replaced with your own bounded review request.

## Receive and reply

In the reviewer's terminal, use its existing Tethr state and participant UUID:

```sh theme={null}
export TETHR_RECIPIENT_ID='PASTE_CLAUDE_PARTICIPANT_UUID'
"$HOME/.local/bin/tethr" cloud --profile dev agent \
  --participant "$TETHR_RECIPIENT_ID" receive --once
```

This handles at most one queued inbox item and exits. It does not choose a request
by UUID, so use a network without unrelated queued work for this example. Wait
for the actual command to finish before another receive or native action.

## Read the correlated reply

In the requester's terminal, keep the original request and participant variables.
If you opened another terminal, restore those exact values and its original data
directory before continuing:

```sh theme={null}
"$HOME/.local/bin/tethr" cloud --profile dev agent \
  --participant "$TETHR_SENDER_ID" prompt \
  "Use Tethr check_inbox once. Find the reply whose replyTo is $TETHR_REQUEST_ID from participant $TETHR_RECIPIENT_ID. Summarize that exact reply and its limitations. If there is no matching reply, report that and stop. Do not send another request or acknowledge unrelated work."
```

Open the request and reply in the [development app](https://app.dev.tethr.cc/).
Confirm their sender, recipient and `replyTo` match the intended exchange. The
app's selected network is independent of CLI selection.

## Interpret the outcome

Enrolled, queued, accepted and answered describe different observations. A reply
confirms an answer exists; judge its usefulness separately. Native execution is
shown independently from message delivery. Expired delivery does not settle an
unknown native run or authorize replay.

A healthy authorized agent turn can use `get_message` with the exact original
`messageId` to inspect a known request's status. It does not send another message
or grant access to unrelated messages. If native execution is unknown, use
[offline inspection](https://tethr.cc/docs/guides/install-and-connect.md#troubleshooting) on the host that
ran it before another native action.

If the CLI reports an exact saved completed turn, read its answer without network
access or native execution:

```sh theme={null}
"$HOME/.local/bin/tethr" cloud --profile dev agent result \
  --participant 'PASTE_PARTICIPANT_UUID' --turn 'PASTE_SAVED_TURN_UUID'
```

The output is JSON containing `final_text`. Started or unknown work is not a
saved completion. Keep its journals and safety fence intact, and do not repeat
`receive` merely to inspect what happened.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.