← Back to concept explainers

Concept Fable

Conversation thread

Conclusion first:A Thread is an object in the hosted Agent API that stores conversation state. It places users, assistants, and messages into the same continuous context, allowing subsequent runs to continue execution based on the existing conversation. In enterprise Agents, the value of a Thread is not simply saving chat records, but isolating the context of a particular customer, ticket, task, or process, avoiding confusion between different sessions, and providing boundaries for auditing, recovery, summarization, and permission control.

Thread · Conversation state object

Read the story first. There is no rush to give a definition here; let the question grow on its own first.

Fable

The thread in the archive that must not be cut

The archive receives many visitors every day. Some come to trace ancestral property, some to pursue debts, some to continue old cases. In the early years, the archivist wrote each inquiry on a separate slip of paper and filed it away once the matter was handled.

An elderly man came three days in a row to look up the same plot of land. On the first day he gave his grandfather's name, on the second he added the boundary, and on the third he asked, "What else was next to that contract from yesterday?" The new librarian only looked at that day's slip of paper and had no idea which contract "yesterday" referred to.

For the first fix, the librarian piled all the slips of paper in front of the old man. The old man was overwhelmed by a dozen irrelevant records, and the librarian couldn't find the context for the current question either.

The second time, they bound by date. The dates were clear, but inquiries from different people on the same day were mixed together. A neighboring merchant's debt case was inserted into an ancestral-property case, and the place name was almost misread.

The old curator took out a red thread and strung together every inquiry, supplementary material, reply, and unresolved question about the same matter. The next time the old man came, the librarian could follow this thread and know what had already been confirmed and what was still missing.

The red lines also had rules. If the old man opened a separate debt issue, a separate line would be started; if an old case ended, the line would be sealed. You could not tangle all conversations into one ball, nor start from a blank slate every time.

The archive later had far fewer misunderstandings. Continuous service requires a traceable thread, retaining relevant context so the next response connects to the previous segment, rather than forcing the visitor back to the starting point.

Later the librarian discovered that the red line also helped them say "I don't know." Looking along the line, if a certain material had never been submitted, the librarian would not pretend to remember it, but would ask the visitor to fill it in. The value of continuous conversation is not just remembering earlier text, but preserving what is known, what is unknown, and the next step within the same matter. The old curator also reminded everyone that when the line gets too long, summaries must be organized, or later arrivals will still get lost. Preserving continuity is not endlessly piling up the past, but catching the prior context useful for the current question. If the line is unbroken, matters do not have to be explained from scratch every time. Later, visitors were also more willing to supply materials, because they knew the librarian would connect them to the right line. This sentence was later written into the newcomers' manual, reminding everyone to look at the on-site cost first before talking about saving effort. And precisely because of this reminder, when similar difficulties arose later, people would not rush to apply old methods.

Reveal

This story is about: conversation threads

A Thread is an object in the hosted Agent API that stores conversation state. It places users, assistants, and messages into the same continuous context, allowing subsequent runs to continue execution based on the existing conversation. In enterprise Agents, the value of a Thread is not simply saving chat records, but isolating the context of a particular customer, ticket, task, or process, avoiding confusion between different sessions, and providing boundaries for auditing, recovery, summarization, and permission control.

What matters is that it is not just a term, but a structural problem that repeatedly appears in real AI / Agent systems. Understanding it lets you judge when to add a model, when to change a process, and when to add governance.

Metaphor mapping

  • Archive: hosted Agent platform or enterprise assistant system
  • Visitor follow-up questions: continuous multi-turn tasks or long-running tickets
  • Wall covered with notes: global context accumulation without session boundaries
  • Reread all the old notes: crudely stuffing in historical messages is costly and prone to confusion
  • Numbered thin lines: Thread objects
  • Notes strung on a timeline: the message sequence within the same conversation
  • Handoffs still connect: different execution runs can reuse the same session state
  • No mixing with others' records: Threads provide context isolation and state boundaries

Andy Ren Xu's judgment

The practical value of this concept is that it helps you break "AI features that look smart" into deliverable, verifiable, and governable units of work.

messagerunassistantconversation statememory
京ICP备2026014272号