Review agent outputs and re-run a single agent
The Agent research tab on any account or contact shows what each agent found, when it found it, the sources it used, and the reasoning behind the answer. Use it after running agents to read the results, judge their quality, give feedback, and re-run a single research agent without re-running everything else. For background on what these agents do, see Agents.
- Open an account or contact, then select the Agent research tab.
- On an account, the page splits results into two cards: Qualification agents (whether the account fits your criteria) and Research agents (the facts and signals that were gathered). On a contact, you see a single Research agents card.
- To jump to one result, use the left sidebar. It lists every agent by name under Qualification Agents and Research Agents; selecting a name scrolls to that result. To narrow the list, type in the Search for name or value… box at the top of the sidebar. It filters by the agent’s name or by the text of its answer, and the cards on the right filter to match. If nothing matches, the page shows that no agents were found for your search term.
- Read a result card. Each card shows:
- The agent’s name and its answer, rendered with formatting. An empty answer shows a dash.
- A timestamp chip showing when the result was last updated.
- A Source and reasoning chip. Hover it to see the Source links the agent used and the Reasoning it gave for its answer. This chip only appears when the agent recorded sources or reasoning.
- Two small meter icons showing the agent’s Effort and Speed setting. Hover each for the exact value.
- To give feedback on a result, use the two buttons on the right of the card. The thumbs-up button records that the answer was helpful and sends right away. The thumbs-down button opens a Share feedback box with the prompt Please tell us why this response wasn’t helpful to you. - type your note and select Submit. Either action shows a confirmation that your feedback was sent.
- To re-run a single research agent with its current settings, select the restart button at the far right of that card. The tooltip reads Research again. This re-runs only that one agent for this record - no other agents and no other records. The button shows a confirmation that research was triggered.
The restart button appears only on Research agents cards. Qualification answers are produced by running qualification, not by re-running one agent at a time. To refresh a qualification result, re-run qualification - see Run a Qualification agent.
Reading an out-of-date result
Section titled “Reading an out-of-date result”If an agent’s settings were changed after this record was processed, its timestamp chip turns orange. Hover it for the exact date the agent was updated, with a note to re-run the agent if you want results based on the latest version. For research agents, the restart button on the same card does exactly that.
On an account, if new Qualification agents were added after the account was last qualified, the Qualification agents card ends with a Missing Qualification agents panel listing them. Re-run qualification to fill those in.
If a result card shows a small orange warning icon instead of the meter icons, the underlying agent could not be found. This usually means the agent was renamed or removed in the Agent Training Center after this result was saved.
What happens next
Section titled “What happens next”Selecting Research again starts a research job for that one agent and record. You can watch it on the record’s Task feed tab. When the job finishes, return to the Agent research tab - the agent’s answer, sources, reasoning, and timestamp update in place, and the orange “out of date” chip clears. Re-running uses credits, the same as running the agent the first time; see Credits.
Feedback you submit is recorded against that specific result and does not change the answer.