Skip to content

v3.21.21

Choose a tag to compare

@dosco dosco released this 20 Nov 20:21
· 7 commits to master since this release

Changelog

  • ba647e3 Merge origin/master: v3.21.20 and the Postgres table-function fix
  • e507f5e Merge remote-tracking branch 'origin/master'
  • 1f7a7de Merge the cross-database SQLite lock fix
  • 7d15eac Update README.md
  • caa63a2 chore(deps): Ax 3ad7b699 to 8dbb33d6
  • e03aa56 chore: bump version to v3.21.21 [skip-release] [skip ci]
  • 7414dcf chore: upgrade Ax Go dependency to 25.4.9
  • 62160c8 docs(site): say what "governed" meant instead of saying "governed"
  • 478f4d2 docs(site): the README and website are DeepORG, built on GraphJin
  • d45fcce docs(site): use the DeepORG name in the homepage testimonial
  • f372be3 feat(agent): continuations, follow-ups, autonomy levels and budgets
  • 8ea635b feat(agent): engine-written display hints for answers
  • a6ae5f1 feat(agent): hold back follow-up questions that lose the user's reference
  • 528579b feat(agent): hold back results that name rows only by their ids
  • 2ae8612 feat(agent): name the rows an answer picks out, not just their ids
  • ee88f53 feat(agent): rate an answer and turn wrong answers into eval drafts
  • 7c27dbe feat(agent): read next, upcoming and soonest as from today on
  • 16edd8e feat(agent): routines propose data changes for their owner's approval
  • 37c322c feat(agent): routines run on the lead, and its reply states the change
  • 8c2c52c feat(agent): rows a watch hands a routine count as evidence
  • 756c615 feat(agent): stream the lead's reply text as it is written
  • 1d5b14f feat(agent): the data agent can be built to read a result before going on
  • d48bb38 feat(agent): the data agent reads saved queries, it never runs them
  • aa3b2b5 feat(agent): the data agent's catalog leaves saved queries out
  • e2dd2c2 feat(agent): the data agent's first prompt lists the values a column holds
  • f04e863 feat(agent): the lead computes derived figures in code and hands them on
  • 53a3077 feat(agent): the lead knows what the caller may change
  • c544d1d feat(agent): the lead reads an answer before it asks again or finishes
  • 2b4e5f6 feat(agent): write approvals hold agent data writes for the user
  • 38861a4 feat(bigquery): connect graphjin serve to BigQuery
  • fdb5c66 feat(catalog): table cards teach names, joins and a database total
  • 3a27674 feat(console): approve or reject a held write from the chat
  • 82661a8 feat(console): draw answer charts from display hints
  • 868baaa feat(eval): mine tests from the questions people already asked
  • 86d269e feat(eval): public suite 2028.5 scores saved metrics on the agent's query
  • 02a69ef feat(serv): OpenAPI description for the agent endpoints
  • 6dfc192 feat(serv): notifications, the inbox GraphJin posts to when you are away
  • 8345e04 feat(serv): routines run on the data agent
  • e8fd868 feat(serv): routines start with their tables' catalog detail
  • ac851f2 feat(serv): service identity for organization agents, with delegation recorded
  • e5e851f feat(site): drop the link grid at the bottom of the homepage
  • 0757ac3 feat(site): redesign the website around DeepORG
  • fac4759 feat(site): say what the training environment is in plain words
  • 251bd3f feat(site): shorten the homepage agent section and use the demo's real accounts
  • 6e92dd7 feat(site): show the benchmark at a glance, with provider marks
  • add92d2 feat(watches): remove agent enrichment; watches reach agents through the lead
  • a2162d0 feat: changes carry their expected outcome, checked later, on a ledger
  • c610866 feat: routine eval, smoke suites and docs for the proactive agent
  • c0a31cb fix(agent): a NOT NULL failure names a missing value, not a missing column
  • c23d802 fix(agent): a blocked lead reply always carries a refusal
  • be02472 fix(agent): a failed query no longer strands the data agent
  • fda137f fix(agent): a lead block its own run explains keeps its own code
  • 6c0f18f fix(agent): a lead clarification is a needs_clarification reply, not an error
  • 331e756 fix(agent): a routine can't claim a change it never made
  • ea2e77a fix(agent): credit a unit conversion when the source names the unit
  • e1d5500 fix(agent): credit figures the lead's code hands on with final() and respond()
  • 291cbbc fix(agent): give the data agent's code blocks at least a minute
  • 2b3aa9d fix(agent): lead callables answer under their advertised names and outlast 5s
  • af78960 fix(catalog): a question's ordinary words no longer outrank what it asks about
  • 3002d41 fix(core): a GraphQL null literal writes SQL NULL, not the string 'null'
  • d989a75 fix(core): a cross-database mutation runs its SQLite script one statement at a time
  • 4a330da fix(core): a refused cross-database mutation runs none of its halves
  • 4d61469 fix(core): a system root mixed with a database table is refused for that
  • e149e46 fix(core): don't report keys the result already names
  • c3c3f25 fix(demo): the demo's SQLite database no longer uses a shared cache
  • ccc37ee fix(eval): agent-data-eval boots its server on --port
  • ffd0367 fix(eval): benchmark test-version labels drop "governed" on the next publish
  • 98b5981 fix(eval): public suite 2028.6 retires the multi-turn family
  • bb68bb5 fix(eval): routine precision reads what the change did
  • a016447 fix(serv): config schema carries core, auth and agent doc comments
  • 3f22253 fix(smoke): restore borrowed tickets newest first
  • 189445d fix(watches): an update drops stored agent enrichment instead of refusing it
  • 6f5929f refactor(agent): one reply contract for the lead and the worker
  • fd6b549 test(bigquery): run core's BigQuery SQL through the driver in CI
  • 3447065 test(clickhouse): skip the sibling-root naming case ClickHouse cannot run
  • adab7c8 test(eval): stop the step bridge idle test depending on how fast the agent starts
  • be571b2 test(smoke): the refusal check asserts the contract, not which guard fired