From fed69b78fadb19277896e6b27eee93634d06ba91 Mon Sep 17 00:00:00 2001 From: rsindall Date: Tue, 4 Aug 2026 01:46:34 +0100 Subject: [PATCH] adr: request Exa connector Product ask for first-class Exa search/enrichment tooling in QM. --- adrs/exa-connector.txt | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) create mode 100644 adrs/exa-connector.txt diff --git a/adrs/exa-connector.txt b/adrs/exa-connector.txt new file mode 100644 index 000000000..b05d2ae93 --- /dev/null +++ b/adrs/exa-connector.txt @@ -0,0 +1,25 @@ +exa connector + +It would be really useful if QM had first-class Exa support for agents — +web / neural search as an Admin-configured connector (or MCP-shaped tool +surface), so orgs don't have to hand-roll sandbox CLIs and API keys. + +What we'd want, roughly: +- Admin can register an Exa API key (org-shared) or per-user key from Keychain + if that fits the model +- agent can run common Exa reads first: search, contents / highlights, find + similar +- clear rate-limit / cost controls so company rollouts don't blow the budget +- works from agent turns without requiring a custom sandbox image just for Exa + +Why this belongs in QM: research agents lean hard on high-quality web search. +Today this can be approximated with an Exa CLI or HTTP calls from the sandbox +plus a skill, but a supported connector would match how other SaaS tools show +up under Admin → Connectors and be much cleaner for rollouts. + +Happy for you to shape this as: +- a first-class HTTP/API connector with built-in Exa tools, or +- a first-class MCP connector recipe for Exa, or +- both + +Not a security report — just a product ask.