Skip to content

Using Telha Day to Day (End Users)

You will meet Telha in two places: the chat (Teams, Slack, or the web app) where you ask questions, and the occasional card where Telha asks YOU something. This guide covers both.

Asking questions

Talk to the Telha bot the way you would ask a well-informed colleague. The difference from a normal chatbot: every answer is generated from your organization's recorded state, and every claim in it is checked against evidence before you see a verdict.

Good questions to ask, and why they work:

"Show me all versions of contract C-1042." Telha keeps every version of every fact, so this returns the full amendment chain with dates: what changed, when it took effect, and when the organization learned about it.

"What changed in our risk assessment between Q1 and Q2?" This runs a temporal compare: two full snapshots of the relevant state, diffed. You get added, removed, and changed facts, not a summary of two documents.

"Find all evidence related to supply chain delays." A hybrid search: semantic similarity plus the knowledge graph plus time filters. Results carry links to the exact passages in the source systems.

"Summarize the current state of Project Alpha with full provenance." A generated answer where every sentence is traceable. Click any claim to see the exact source text it rests on.

Time-travel phrasing that Telha understands well:

  • "as of last March" (what was true then)
  • "what did we know on Feb 1 about X" (what we believed then, before later corrections)
  • "before the amendment" / "at the time we signed"

Reading an answer

Three things to look at:

  1. Tool chips. While Telha works you see which tools ran (search, history, compare). This is transparency, not decoration: an answer's shape follows from what was looked up.
  2. Verification badges. Each claim in a generated answer is marked supported, partial, or unsupported. Unsupported does not always mean wrong; it means the evidence Telha planned does not back that sentence. Depending on your workspace policy, unsupported text is either flagged or removed.
  3. Citations and the trace. Numbered markers like [S3] link claims to sources. "View trace" opens the full evidence trail: which spans were selected, which were rejected for budget, and how each claim scored. The deep link opens the original document in its home system, behind that system's own login.

Caveats matter. If an answer says a fact is "pending clarification", Telha is telling you a date or value in it is a best guess currently being confirmed with a colleague. The answer names who was asked. Treat it accordingly, or wait for the confirmation.

When Telha asks you something

Sometimes a card arrives in your DMs: "Which date should apply to this contract clause?" with two or three buttons like "12 June (signature date)" and "27 June (countersignature)", plus a "Something else…" option.

What is happening: a document came in with an ambiguous date, and the graph says you are the person best placed to resolve it (you wrote it, you are named in it, or you own it). What you should know:

  • One tap is all it takes. Your answer is recorded as evidence with your name, your role, and the moment you answered. This is a real correction to company state, not a poll.
  • High-stakes facts need two people. For critical dates (expiry, liability, obligations) the card says so, and the correction only binds after a second authorized person confirms. You will see "First confirmation recorded" until then.
  • Disagreement is handled, not hidden. If someone else answers differently, both answers are kept, the fact stays uncorrected, and a steward resolves it. You may change your own answer any time before it binds.
  • "Something else…" lets you type the right date if none of the options fit. It must be an actual date; Telha will not guess from prose.
  • Typed replies do not count. Only the card buttons record answers. If you reply in free text, the bot re-prompts with the card. This is deliberate: casual chat must never silently become evidence.
  • You will not be spammed. There is a daily cap per person, and multiple questions batch into a single digest.
  • If you should not see the underlying text, you will never get the card. Telha checks read permission before every delivery.

Asking Telha in chat, scoped to you

You can also just ask the same bot questions ("what is the history of contract C-1042?"). Answers are scoped to YOUR role and permissions: two colleagues asking the same question can legitimately get different answers, because Telha answers from what each is allowed to see.

What Telha will not do

  • It will not show you content your role does not permit, even in an answer's citations.
  • It will not present a guess as a confirmed fact; guesses carry caveats.
  • It will not let a chat display name act as an identity. Binding actions require your verified corporate identity; if yours is not linked yet, you will get a message saying so rather than a silent failure.