v0.1.19

Update dependencies to latest.

v0.1.18

Feedback grew up. Votes can carry a comment explaining them, from the CLI or a note field beside the web vote buttons. whodar feedback list shows everything the ranking has been taught, and whodar feedback clear undoes it by question, person, channel, or entirely. A new --feedback flag on ask, serve, and bot tunes how hard votes move ranking, from off through high. Recording is now the record subcommand.

v0.1.17

whodar can now use Claude and OpenAI, behind the egress policy that has always guarded it. The strict default still refuses anything cloud. Opt in with the redacted policy and candidates leave as anonymized numbered roles, no names, emails, or message text, with the model returning numbers and the summary written locally. The openai provider speaks the common chat-completions format, so it also reaches local servers like LM Studio with no opt-in needed. Keys come only from the environment.

v0.1.16

Picking llm mode without Ollama running now explains what to do instead of dumping a dial error: install Ollama, pull a model, ask again. Keyword mode always works.

v0.1.15

Try whodar in one command: whodar demo builds a simulated company across all eight sources and opens the web UI over it, no credentials needed. Clicking a person now opens their full profile: email, joined identifiers, manager, channels, and everything they know, deep-linkable with ?person=. Also fixes vote buttons scattering across result cards.

v0.1.14

Clicking a person now shows who they are instead of launching a mail client: email with copy and an explicit email action, joined identifiers, and their strongest topics. A single-result answer arrives expanded, and answer JSON gains a topics field.

v0.1.13

The README is now a front door instead of a manual: pitch, demo, install, quickstart, and pointers. The detail lives in docs/REFERENCE.md, which covers every command, flag, source, credential, and file, and in the help text itself: whodar index --help now lists all eight sources with their scope flags and environment variables.

v0.1.12

The README now shows whodar working: a terminal recording of asking and voting, and a web UI screenshot with confidence badges and feedback buttons. Queries in the web UI are shareable links; open one and it runs.

v0.1.11

whodar now tests itself against a simulated company. Mock servers speak the real wire formats of Slack, GitHub, Jira, Confluence, and PagerDuty, alongside a real git repository, an org chart, and CODEOWNERS, so the whole pipeline runs end to end with no credentials. The GitHub source also stops counting bot accounts as people.

v0.1.10

whodar now learns from its users. Tell it a result was right or wrong, from the new whodar feedback command or the buttons on every web result, and future rankings for that question move accordingly. Votes are capped so they tune answers without burying the evidence, survive re-indexing in their own file, and results say when feedback boosted or lowered them.