Geiger Grey · Cited answers
Answers you can check
Every claim points to its source
Ask in the Playground and each sentence carries a numbered citation. Open one to read the passage, or see every passage the model was sent and how well it matched.
Geiger GreySee exactly what the model was given
Each reply records the passages it was sent, with their retrieval scores, token counts and which ones it cited. Click a citation to read the passage and open the source.
Keep it to your own sources
Knowledge base only is on by default in the Playground. Turn it off and the model may add general knowledge, and it says when it does.
Try a setting without saving it
Run settings show the pipeline's model, search, threshold, rerank, graph and citation settings. Override the temperature or the number of passages for this tab only.
Citations land on the page or the second
A citation names the page of a document or the moment in a recording. Open a cited video and it starts where the passage begins.
Every conversation, from every channel
Conversations keeps the chats from the Playground, the widget and the API, each with its transcript, citations and feedback.
Tune how it finds the answer
A pipeline holds the model, the knowledge bases, the retrieval settings and the prompt. Change it, try it in the Playground, then publish it as a new live version.
Geiger GreyRetrieval settings you can read at a glance
Set how many passages to fetch, the similarity threshold, hybrid search, reranking and citations, and when to consult the knowledge graph: off, auto or always. Publishing snapshots the configuration as a new version and makes it live.
Drafts stay out of the way
A pipeline stays a draft until you publish it, and each one answers from the knowledge bases and model you chose for it.
A system prompt in your own words
Write the instruction every answer is grounded with, and place the retrieved passages and the question where you want them.
Measure it before people rely on it
Score a pipeline against your own questions, triage the thumbs-downs and unanswered questions, and trace any request back to the passages it used.
Geiger GreyEvaluations on four measures
Build eval sets of questions with expected answers and sources, then run them against a pipeline. Each run is scored for faithfulness, answer relevance, context precision and context recall.
Turn a thumbs-down into a test
Feedback gathers negative and unanswered replies. Open the conversation, resolve it, or add it to an eval set so the next run checks it.
Trace any request
Request Logs covers the Playground, the widget, the API and MCP. Open a request to see the chunks it retrieved, how each was found and its score.