Connect Jira, Confluence, Notion or Linear and Khint keeps a local, always-fresh index of your work. Your agents answer with your real tickets and pages, and nothing is copied to our servers.
PAY-482 is blocked on the ledger migration. Per the retry policy RFC, soft declines now move to a 72-hour window with exponential backoff, and the migration lands on the 14th.
A model that has never read your work writes the paragraph anyone could have written. The gap isn't intelligence, it's access: every fact on the right came out of a document Khint found locally before the run started, and nothing was pasted or uploaded to get it there.
An index is only trustworthy if you know what's in it. Four tools, a defined slice of each, a pass that only fetches what changed, and a hard ceiling on what any single run is allowed to carry.
Connect Jira, Confluence, Notion or Linear and each one gets its own toggle. Khint walks it with your credentials and writes what it finds into a full-text index in a SQLite file on your Mac. Up to 2,000 documents per tool, the oldest dropped as newer ones arrive.
After the first pass, only documents that changed since the last one are fetched.
A run doesn't ship your index. It ships the three best-matching excerpts for the text you launched it on, and none when nothing matches well.
There's no embeddings service to pay for and no vector database to run. Indexing makes no AI call at all, and retrieval is ranked full-text search against a local file, which is why it answers in milliseconds and why switching it on doesn't change your bill.
There's no upload step, because there's no destination. Khint reads your tools with your own credentials and writes the index beside your other local data, and a run carries excerpts rather than access.
Security and privacyEnterprise AI platforms sync a copy of your corpus to their cloud and search it there. Khint inverts the model and leaves the corpus where it already is.
Connect the tool, switch the source on, and every agent reads it from the next run.
Add your Jira, Confluence, Notion or Linear credentials once, on Khint's Integrations page.
Knowledge sits under the session list in Memory. One toggle per tool starts the index, and it is the only switch there is.
See MemoryNothing to set on the agent. From the next run on, every agent searches the index for the text you launched it on and reads the excerpts that match.
See what an agent readsKnowledge is what your agents know. These three are what they do with it.
On your Mac, in a local SQLite full-text index, never on Khint's servers. Cloud AI platforms keep a synced copy of your Confluence or your CRM on their own infrastructure and search it there; Khint holds no copy of your corpus anywhere. Switching a source off, or forgetting an integration's credentials, deletes its local index immediately.
Only the three excerpts that best match the text of the run you launched, and none when nothing matches well: a few hundred words, sent inside a tagged data block, the same way your active session context travels today. The index itself never leaves your machine, retrieval is pure local search with no AI call, and the excerpts are read as data rather than as instructions, so a document can't tell your agent what to do.
Jira (issue key, summary, status, description), Confluence (page title and body, per space), Notion (page title and block content) and Linear (identifier, title, state, description). Each toggles independently from Memory's Knowledge section. The first pass fills the index in the background; after that only documents that changed since the last pass are fetched, every 30 minutes or on demand with Sync now.
No. Indexing makes no AI calls at all, and retrieval is local BM25 full-text search that answers in milliseconds. There's no embeddings service to pay for and no vector database to run. The only cost is the few hundred extra words the excerpts add to a run that found a match, and the normal per-run rounding usually absorbs them.
Khint indexes with each user's own credentials, so every person's index contains exactly what their Jira or Confluence account can already see, and nothing more. There's no central copy and no access-control re-mapping to get wrong. An index is also never carried into a shared team pack or a starter pack: what your colleague clones is the agent, not your documents.
Toggling a source off deletes its local copy on the spot, and so does forgetting that integration's credentials. Deleting your Khint account purges every indexed source. Signing out on its own keeps the index, exactly as it keeps your local history, so coming back doesn't mean a full re-sync.
Free with 300 credits a month. No credit card. macOS 13+ or Windows 10 22H2+.
