History¶
Ask an agent to find something you discussed earlier, whether it happened in this channel, another linked channel or one of your collected native sessions.
Find our discussion about the launch checklist and remind me what we decided.
Find an earlier conversation¶
Give the agent a few distinctive words from the topic. Ask it to read the surrounding messages before summarising a decision, so a search hit is not mistaken for the final answer.
Search for “release checklist”, then read the messages around the result. What was still unfinished?
You can also open the thread in Trunkline to read it yourself. If a result includes a message reference, reuse that reference when asking the agent to look more closely at the conversation.
Bring history across channels¶
Link your channel profiles to the same account. Your agents can then search conversations you participated in through those profiles. Native sessions join that history when their client is connected to your account.
Without linking, a channel identity has access only to the history associated with that identity. If the agent cannot find a conversation from elsewhere, check the profile link and whether the session appears in Trunkline.
What you can find¶
Search covers visible message text: your prompts and the replies in recorded conversations. It does not search inside attachments, tool arguments or the agent's private reasoning.
Use short, distinctive English terms. Every term must match, so adding words can make a search too narrow. Try fewer words or terms you remember using if nothing appears. Search does not understand meaning in the way a model does; it will not reliably find a conversation from a vague description or a phrase in another language.
Which conversations are available¶
An agent answering you can read recorded threads where one of your linked profiles has spoken. That includes other participants' messages in those threads. It does not include every conversation in a group merely because you are a member.
If important context is missing, provide it in the current conversation rather than assuming the agent has seen it.