● all systems operationalv0.17.10
stech/

Persisted conversations

The agent detail page's "test the agent" panel keeps history. Each chat is a named conversation, scoped per user, that you can switch between, rename, or delete. Replaces the old stateless prompt box where every turn started fresh.

Layout #

The agent detail page is split into two columns:

  • Right pane — full-height chat for the active conversation. Input is sticky at the bottom; assistant turns scroll above. Tool calls render inline as collapsible cards within the assistant's turn.
  • Left rail — your conversations for this deployment, most-recent first. + new chat at the top creates an empty conversation.

Empty state on first visit: a single big input that creates the first conversation when you submit.

Titles #

Auto-derived from your first prompt, truncated to ~60 characters. You can rename inline from the rail (click the title; Enter to save, blur to cancel). Renaming only changes the display label — it doesn't re-key the conversation.

Switching, renaming, deleting #

  • Switch — click any row in the rail. The right pane reloads with that conversation's history.
  • Rename — inline from the row.
  • Delete — confirm prompt; deletes the conversation and cascades its messages.

Streaming #

Assistant text streams in token-by-token as the model writes it (epic #179). Tool calls narrate inline as they fire. Final usage stats (iterations, tokens, stop reason) land on the message once the run completes and get persisted to the database.

Schema #

Two tables, both ON DELETE CASCADE from the parent so deleting a deployment / org / user cleans up automatically.

  • agent_conversations — id, organization_id, deployment_id, user_id, title, created_at, updated_at. Indexed on (user_id, deployment_id, updated_at DESC) for the rail's most-recent-first query.
  • agent_messages — id, conversation_id, role (user | assistant), content (jsonb, mirrors the SDK's Message['content'] shape), plus assistant-only telemetry: stop_reason, iterations, input_tokens, output_tokens. Indexed on (conversation_id, created_at).

Privacy #

Conversations are per-user. Org-mates in the same organization running the same agent see their own conversations only — your chat history is yours. Sharing a conversation across users (read-only or collaborative) is a separate epic; not shipped.

edit this page on github →