Repository navigation
✨ Customers: goal and satisfaction suggestions with the meetings model (opt-in) - #18806
Conversation
…l (opt-in) PR 9 of RFC #18742. - pkg/customers: Infer reads one customer's cleaned, fenced evidence through a Completer and files suggestions only. A confirmed or manual value is never overwritten, a rejected suggestion is not offered again, there is no numeric score, the reply is bounded and typed, and a person is not read again within 24 hours or while their evidence is unchanged. - cmd/hive: the customers_inference optional sweep on the meetings model route, a daily token ledger of its own, at most three people per pass, audited without text. - customers.inference.enabled, default off, with a Features switch. Refs #18742 Signed-off-by: Andy Anderson <andy@clubanderson.com>
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
Refs #18742 Signed-off-by: Andy Anderson <andy@clubanderson.com>
Mutation testinggremlins found 10 surviving mutant(s) in changed files: the code ran under tests, but changing it did not fail any test.
13 mutant(s) in changed files are not reached by any test (NOT COVERED). A maintainer may apply the
Verdict: |
|
[APPROVALNOTIFIER] This PR is APPROVED Approval requirements bypassed by manually added approval. This pull-request has been approved by: The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
PR 9 of the Customers series (RFC #18742), the last in the plan. It builds on PR 7's
SafeGoal, which is already merged.The owner asked for goals and satisfaction that hive infers and offers for confirmation. The data model, the provenance and the profile UI (confirm, edit, reject) were built in PRs 1 and 3. This adds the inference.
Decision for the owner: it is OFF by default
customers.inference.enableddefaults to false (envHIVE_CUSTOMERS_INFERENCE_ENABLED; a switch in Settings, Features). Unlike the rest of Customers, it sends text to a model: a customer's own issues, pull requests and approved mail go to the model configured for meetings. That text can be about people who never signed up, so I did not make it default on. Say the word and it flips; it is one default. It also needscustomers.enabledand a meetings model that is an OpenAI-compatible endpoint or the Anthropic API.What it does
customers_inferenceoptional sweep, every 30 minutes, makes no GitHub call. It reads at most three due customers per pass, only customers (a suggested person is never read), and leaves a person alone for 24 hours after a read and while their evidence is unchanged.meetllmthroughbetty.MeetLLM), the one the persona uses. Charged to a daily usage ledger of its own (kept seven days) with a 40,000 token budget; a spent budget ends the pass. An agent backend is not used.Rules, each tested
ioscanfinding, fenced with a fresh nonce; the system prompt says it is dataTestSafeEvidence,TestInferencePromptFences...,TestInferencePromptCarriesNoPrivateFields...TestParseInferenceBoundsAndTypesTheReplyTestInferenceNeverOverwritesWhatAPersonOwnsTestInferSkipsAndFailsSafelyTestRunCustomersInferenceLimitsAndStopsOnBudget,TestCustomersEvidence...TestCustomersInferenceDefaultsOffAndNeedsCustomers,TestGovernorFeatures_CustomersInferenceToggleRoundTripAudit:
customer_inferencewith the person id, goal count and whether a satisfaction suggestion was filed. Never text. Nothing inferred leaves the hive.Docs
Design page (status, signals, rules),
security-model.md(the egress and what is done about it), operator reference, env vars, example config, changelog.Refs #18742
🤖 Generated with Claude Code