0.8.0: Hosted AI with owner approval

@andersalm

Release: 0.8.0 (stretch, Mac only; cutoff Thu 22 Oct 18:00 UTC). Without both review PASSes and the installed proof by the cutoff, 0.8.0 ships the fallback: existing hosted connections stay visible and removable, and new hosted connections and hosted sharing are blocked.

Owner: Anders (@andersalm). Platform scope: Apple Silicon Mac only (decision, 19 Oct). Linux and Jetson are not in the 0.8.0 claim.

Done means

Release acceptance: Each exposed feature works clearly from start to end, including recovery. Keep unfinished or confusing features hidden until their complete journey passes. Hiding a feature does not complete its task; source checks alone do not establish installed acceptance.

The owner adds an OpenRouter or Venice connection, approves its exact
internet access in Inbox, and receives an Assistant reply.
Runtime owns that authority across every supported request path. End
prevents further calls; the result of a concurrent reply is reported clearly.
Removing the temporary operator route preserves saved connections and data.
One bounded real request per provider and installed authority/isolation
checks pass on every platform named in the release claim.

Source-Home proof so far (c4074b463, not a release): Venice, OpenRouter. Candidate: see #93.

Remaining checks (installed, Mac). Checks on the test Home at the current train head are pre-release evidence; alpha.25 installed acceptance, including the update from alpha.24, and Irzhy's review before stable stay required:

A wrong OpenRouter key is rejected at Check key.
System refreshes live after an Inbox decision.
Saved connections keep working after updating from alpha.24.
Installed authority and isolation checks pass on the Mac (box 4).
Human review by Irzhy (by Wed 28 Oct, listed in 0.8.0: Release 0.8.0 #93).

Jev remains an optional, disabled-by-default extension.

Keep this hosted lane separate from #84's local/offline acceptance. Reuse the typed model operations and explicit Runtime route/grant ownership. Local capacity/refusal never silently selects a hosted provider.

Retained ordinary-use cases: request owner consent before key validation; retries reuse one pending approval; test Approve → validate/save → select → reply, restart, then End or expiry followed by a new request. Exercise cancelled and failed End operations and concurrent replies. Distinguish cancellation, timeout and uncertain outcome in the user flow.

The final operator-selected saved fix is 1825dc4b: coordinate hosted End with Runtime send admission across Inbox and System.
Use approved base/target feat/0.7.1-models@ea670e27; the pick contains six code/test files and no historical notes.
The saved candidate has one availability gap: cancelled or failed End after signaling can interrupt an admitted send before commit. Keep this distinct from committed End and from installed acceptance.
Next: apply the fix, run focused send/End and Inbox/System tests, and review the authority/cancellation boundary. Real-provider and visible Assistant acceptance remain open.

The saved End/send fix is now a six-file cherry-pick on ea670e27.
Its prompt-only native probe conflicted with the base's conversation tests. The resolution reuses #141's accepted native v1/v2 fixture and retains exact conversation, role/order and token-cap assertions.
Production send/End coordination remains the saved concern. Next: focused egress and Inbox/System tests, basic gates, and independent cancellation/authority review.
The cancelled/failed-End availability case and installed real-provider/Assistant acceptance remain open.

Candidate b821874e passes all four basic gates and 25 focused egress tests, including native v1/v2 conversation probes.
Independent review found two source gaps: End signals admitted work before its durable write, and System End/delete keeps an earlier admin check across new waits.
The correction will commit authority first, then drain successful scopes in operator-before-offer order, with current owner checks after waits.
Next: add production cancellation/failure and revoked-owner regressions, apply the correction as a separate commit, and repeat the affected checks.
Installed concurrent Assistant/real-provider proof remains open. The inherited cross-Home job-reservation collision is retained here as a separate finding.

The source correction now places Inbox/System End signals after durable authority writes, with operator-before-offer drain order and partial operator failure shown explicitly.
System End/delete rechecks current admin authority after setup/transition waits; deletion checks again after the model-sharing wait.
New regressions exercise a real loopback validation send during cancelled End and a forced decision-write failure, plus revoked-admin End/delete across each wait.
Next: compile and run these regressions, repeat the affected tests, and review the corrected candidate before publication.

Candidate b821874 plus the current correction passes the real loopback End regression: cancelled End leaves saved authority ended; a failed authority write leaves the admitted send active.
The revoked-admin regression passes for System End/delete after setup waits and delete after the sharing wait.
Correction diff SHA-256: 8fd6c63e408cb6df9a57f4e06205bd9248fe9dad88c2a39745ab13d0afb7a184. Independent security review is in progress.
Next: finish affected tests and basic gates, resolve review findings, then commit and publish a draft PR. Installed provider journeys remain open.

Security review found two remaining paths: staged discard used authority checked before its wait; delete/discard could save End and then skip send notification during later waits or cleanup failures.
The correction now rechecks staged-discard authority after setup/transition waits and drains each committed End before later waits.
Offer removal separates the saved configuration change from credential cleanup and provider refresh, so committed removal signals End before either can fail.
The earlier 25 egress tests, four Inbox tests, staged-discard test and basic gates passed. The expanded revoked-admin test is compiling.
Next: verify the remaining paths, obtain security review of the final diff, then publish the draft candidate. Installed acceptance remains open.

The expanded revoked-admin test passes all seven setup, transition and sharing cases. The delete/discard drain test passes alone; its first parallel run exposed a test timing race, which now waits for the saved change.
Review also found that zero-decision delete needs a current-admin check after its drain, and stale staged discard must check key presence before End.
The delete check and revocation regression are added; the staged-presence correction is next. Candidate remains local and under review.
The inherited global job-create key omits Home identity; this separate finding remains open in #85.
Next: complete those regressions and final security review before publication. Installed acceptance remains open.

Prepared candidate 98b4acc (tree recorded with the PR), with saved fix b821874 and a separate durable-End correction; both commits target feat/0.7.1-models at ea670e2.
Tests pass: 48 System cases, five staged cases, real loopback End cancellation/write failure, 25 egress cases and 10 hosted-config cases; all four basic gates pass.
Security review passes the scoped End correction: full candidate diff SHA-256 3552839624fa8af98faf46e31365570aec1829d8dcbd5ed80b36c0ce7f3099e1.
Before publication: the new task branch is two commits ahead and zero behind origin/feat/0.7.1-models; the worktree is clean. I will push to origin/fix/0.7.1-hosted-end-admission and open a draft PR.
Installed provider journeys and the inherited cross-Home job-create key gap remain open; Irzhy will be requested when the PR is ready.

Prepared candidate 98b4acc, with saved fix b821874 and a separate durable-End correction, targets feat/0.7.1-models at ea670e2.
Tests pass: 48 System cases, five staged cases, real loopback End cancellation/write failure, 25 egress cases and 10 hosted-config cases; all four basic gates pass.
Security review passes the scoped End correction: full candidate diff SHA-256 3552839624fa8af98faf46e31365570aec1829d8dcbd5ed80b36c0ce7f3099e1.
Before publication: the new task branch is two commits ahead and zero behind origin/feat/0.7.1-models; the worktree is clean. I will push to origin/fix/0.7.1-hosted-end-admission and open a draft PR.
Installed provider journeys and the inherited cross-Home job-create key gap remain open; Irzhy will be requested when the PR is ready.

Published draft PR #148: #148, commits b821874 and 98b4acc, targeting feat/0.7.1-models.
Passing source proof: 48 System cases, five staged cases, real loopback End cancellation/write failure, 25 egress cases, 10 hosted-config cases and all four basic gates.
Independent security review passes the scoped End correction, full candidate diff SHA-256 3552839624fa8af98faf46e31365570aec1829d8dcbd5ed80b36c0ce7f3099e1.
Open: CI for this candidate, installed OpenRouter/Venice journeys, platform acceptance, the inherited cross-Home job-create key gap, and Irzhy's review when ready.
Both commits are protected by origin/fix/0.7.1-hosted-end-admission; the worktree is clean and local/remote divergence is 0/0.
At the operator's request, work stops until the weekly reset. LA-03 resumes after that reset; this issue remains open.

Completed-result triage for draft PR #148, candidate 98b4acc9eee0d4e284d122455505056a0640ace0, tree a23fec1f5b9bdabff1274ec894841c9610b635e1: CI run 36767214431 has finished.
source-gate and custody-harness-smoke pass. Lint, both test jobs and all three source-home jobs fail; release publication is skipped.
Linux builds share E0433 errors: Unix callers use api::model_provider_egress, whose module declaration is gated to macOS. Mac source-home fails -D warnings on the unused remove_hosted_offer import and function.
The capsule suite also fails private_request_write_failure_kills_and_reaps_child at supervisor.rs:819 (result.is_err(); 92 cases pass, one fails). That is a separate failure to review.
Corrections and passing candidate CI remain open. Installed OpenRouter/Venice journeys, platform acceptance and the inherited cross-Home job-create key gap still need proof. Irzhy's required review follows draft readiness.
This was read-only log and source triage. Source is unchanged, the checkout is clean, and task-remote divergence is 0/0; PR #148 remains draft.

The a1 remote allocation stopped before checks with the service's "not supported" result, reported by the helper as machine capacity. The a1 Rust controls remain unexecuted; the b4 complete-gate PASS remains qualified predecessor proof.
Candidate a1f7628aa7ce38c3c4ced074043699a1c8549825 is clean and has a passing CI prediction. A later retry will use a fresh owned resource; the writer contract remains unchanged for security disposition.
Installed Mac acceptance and required human review remain open. Publication stays held with its configured hook intact.

The fresh remote retry has allocated an owned resource and is checking exact candidate a1f7628aa7ce38c3c4ced074043699a1c8549825. The earlier allocation refusal created no test result.
A1 CI prediction and the bounded Mac OS fault mechanisms pass; its Rust fault controls are still pending. B4 retains its complete remote-gate PASS.
The writer's uncertainty result awaits security disposition. Installed Mac acceptance and required human review remain open; local builds and publication remain held.

A1's remote behavior suite reports 1,739 passes and zero failures; the b4 run reported 1,740. The JS source is unchanged. Log comparison finds one fewer report for the identically named afterInput unload cases; the unchanged file passes all 70 cases in an isolated Mac run.
The report-count difference is recorded for test-harness triage, with cause still unknown. A1's source/Rust gate continues; its isolated writer controls remain pending.
Writer semantics await OMP security disposition. Installed Mac acceptance and required human review remain open; publication stays held.

Candidate a1f7628aa7ce38c3c4ced074043699a1c8549825 (tree de857656b50dd5519d317662b4eb44b71a498a6e) passes the complete remote gate: current scripts/test-source-gate.sh, required format checks, workspace all-target checks, Clippy, 1,739 behavior passes and 3,663 Rust passes (23 ignored). The gate took 25m10s and its test resource was removed.
The isolated pre-commit refusal and post-rename durability controls pass. The first preserves admitted sends; the second confirms that a returned error can follow visible replacement. OMP owns the writer-semantics decision; the earlier behavior report-count qualification remains recorded for harness triage.
The clean branch stays local for OMP's exact-head Claude/security review and integration. Mac confinement/request checks, installed OpenRouter/Venice replies, artifact parity, manual UX and irzhywau review remain open; Linux stays paused.

Claude cross-family review of fix/85-hosted-ai-080 at a1f7628aa: PASS WITH CHANGES.

P2, fix before merge (requested from the builder): both End paths wait on the per-Home setup lock, so a running key check or Save delays End and its cancel. The fix drops the setup lock from End and keeps the transition lock.
P3, after 0.8.0: after a post-rename fsync failure, End can be in effect while reported as failed, and it cannot be retried. Wanted: a distinct "saved, not confirmed on disk" outcome treated as saved, plus a retry of the directory fsync.
No path reaches a hosted provider without the owner's per-connection approval: the operator route is removed and saved offers are kept.

Candidate f37aa1219bacf16695d75971815c254a3eb4ce7f passes the exact remote gate: 3,664 Rust tests pass with 23 ignored; the behavior report has 1,691 passes/zero failures. The focused End regression passes in 0.16 s; restoring the four old waits reproduces the one-second timeout. The control clone and owned build resource were removed.
The b5847af0 gate failed only the new test's JSON parser; its control was skipped and resource removed. The earlier 1cd5bc02 run was stopped before a completed gate result or control, and its resource was removed.
OMP owns final review and integration. The 49-case behavior report-count difference from b5847af0 with unchanged JS, installed Mac provider replies, confinement, artifact parity, manual UX and irzhywau review remain open; Linux stays paused.

Claude delta review: PASS at f37aa1219 (range a1f7628aa..f37aa1219: 1cd5bc022, b5847af0e, f37aa1219).

The prior P2 is closed. Inbox End (gateway_inbox.rs:506-508) and System revoke (gateway_home_system_ai_provider.rs:580-583) take only the transition gate. The authority rechecks after the wait are intact.
- No lock cycle. The End/Validate race is covered by current_grant_for and ended.changed().
- The regression test is meaningful: with the old waits restored, the control fails (Elapsed at model.rs:2347).
P3, optional, queued to the builder: when End cancels an in-flight key check, the owner sees "could not check the hosted connection. Try again." instead of the ended message. The fix is to map "hosted access ended during send" and "hosted access is ending" to HOSTED_CONNECTION_ENDED, and to remove the dead "owner ended hosted HTTPS" arm.
Test-count gap (1740 vs 1691 on identical JS): the lost results are the ends of two test files. This is consistent with the open Node bug test_runner: --test-force-exit with concurrency silently loses test verdicts (parent reports fewer tests than ran, exit 0) nodejs/node#64833 under --test-force-exit. It was not reproduced locally, and this delta did not cause it. It is tracked separately as a CI reporting item.

Integrated on the rc10 train: be2362dbe..62b919247 (cherry-picks of 3beb81643..f37aa1219). The installed Mac proof with real test keys is still open (cutoff Thu 22, 18:00 UTC). No user-facing 0.8.0 note until then.

Claude delta review PASS covers normal b46d626fea0e70d17210ee876a012fddd46a4299 / tree 05547818bbdd4ff0dce07bb88a9707164ccb6401, above f37aa1219: the ended-access response and regression expectation agree. The reviewer ran 34 System Node tests; Cargo and installed provider behavior remain separate evidence.
The completed current-head run 183211 binds that source, reports 1,602 behavior passes and begins source checks, then ends with a missing remote exit status. It reaches no Rust test phase or terminal gate verdict; the prior f37aa1219 gate pass covers its earlier source.
OMP owns the next admitted completed check and integration. Required human review plus installed Mac provider replies, confinement, artifact parity and the complete setup, share and End journey remain open.

The Hosted AI corrections at c16977b48 passed their configured Linux source gate and hosted2 review; eight matching patches are integrated through 1bcd10ad7. OMP reports further source security reviews of six later Hosted changes and the Inbox notification patch. The Hosted review used selected worktree functions; the notification review used an exported diff and worktree checks. OMP keeps the findings private, has integrated the two reviewed private corrections and owns the remaining Hosted follow-ups and final-head reviews.
The panel correction at 6d7634cc0 passed its configured Linux source gate and hostedcopy2 delta review, which closed the blocking finding at f244feeae. OMP integrated two matching patches through c4074b463, which passed the configured Linux train gate in 27m08s.
Anders confirms that OpenRouter works and previously reported Venice refusal after End and a reply stopped during streaming. OMP records these on the c4074b463 source Home (OpenRouter, Venice); upstream request counts and Venice-specific key coverage remain unverified. The stale End-control port 7823a7334 passed its combined configured Linux source gate in 35m35s. Train predecessor eaf6c6f88 added the reviewed #452 documentation and matching Assistant Copy patches. Copy's blocking source finding at 00d262670 is closed by copy2 delta review at e30dc65ca, whose configured Linux source gate passed in 30m40s. The exact eaf6c6f88 combined configured Linux source gate passed in 24m28s. OMP reports that the recipe's 12 browser checks, run one by one, and all 1,784 behaviour tests passed on Mac. The exact head also passed the full configured Linux source check: gate 24m34s, Runtime tests 9m31s and capsule tests 5m45s (report). Linux CI, package and broader candidate qualification remain open. OMP reported the same source Home rebuilt at eaf6c6f88, with the account and saved connections kept, and sent the existing wrong-key, live End-control and Copy checks. OMP now reports that test Home stopped and removed. Its separate isolated source-Home report at abddf200e supplies partial route-removal and isolation evidence using test providers; real-provider replies and complete candidate acceptance remain open. OMP reports that one test Home at 26f7c7589 has been rebuilt with the supported llama setup and started for the combined Browser and Hosted check. Anders marks the new-account and Hosted parts of the combined source-Home check OK. After a further preparation failure, OMP reports that it repaired the Browser configuration, updated the source installation receipt and restarted the test Home. Anders then reports Browser PASS, with slow YouTube loading. This is human evidence from the modified source Home; exact packaged installation/update, independent artifact parity and the real-provider continuity chain remain open. Anders now reports that Copy works and System updates after reopening. He reports that the Cancel control changes after a key check and that Inbox notifications are intermittent. OMP owns the follow-up checks; the displayed wrong-key result and independent installed acceptance remain open. Inbox correction 98131d9d0 passed notify1 source review and its configured Linux source gate in 24m33s. Form correction b99d9f019 passed form1 source review and its configured Linux source gate in 39m38s. The recovery-fixture delta cb69be4c3 passed notify2 source review. OMP integrated matching patches as 75fcc24ff, 429038dee and 42c054ef7. The exact combined configured Linux source gate at 42c054ef7 passed in 25m42s. Reload guidance correction 1d92e7bec passed reload1 source review and its configured Linux source gate in 27m54s; matching port 31348f25c also passed its full configured Linux source check: gate 22m34s, Runtime tests 9m46s and capsule tests 6m54s. Hosted follow-up d4bb902a3 passed form2 source review and its configured Linux source gate in 24m57s; matching port 5a4386d94 preserves its parent, tree and reviewed bytes. The exact combined configured Linux source gate at 5a4386d94 passed in 22m46s. Train predecessor 26f7c7589 has matching Browser ports a240c4362 and b748597aa and notes9 source review PASS. Its combined configured Linux source gate passed in 25m22s. OMP reports all four Mac source-gate sections, the recipe's 12 browser commands and 1,792 behaviour tests passed on this head. The exact head also passed the full configured Linux source check: gate 22m05s, Runtime tests 9m45s and capsule tests 5m21s (owner report). OMP has now integrated the F3 hint pair as 6442a4f85 and 931959bc6, with the same reviewed changes, and the #89 test commits through ba224ca2b. The configured Linux source gate at ba224ca2b failed in 3m37s in the Home reload test. The configured Linux source gate on 9382cd50e passed in 24m16s. reloadsmoke2 source review is PASS WITH CHANGES; the new test-only commit a6b52a808 applies its remaining null-safe predicate correction. The executed gate covers 9382cd50e; final qualification of the resulting train remains open. Final train qualification and the actual alpha.24 update hop remain open. The contingency's first review failed at 00a2a87ee; corrected c710537bf passed fallback2 whole-branch source review and awaits source-gate and installed qualification. Complete Owner Inbox/connection journeys, exact CI/package checks, installed authority/isolation and required reviews remain open. At 13:01 UTC on 22 October, Anders approved Hosted AI in alpha.25, answering OMP's explicit request for an exception to the 18:00 UTC proof deadline for alpha.25 only (decision). This supersedes the pending-answer record. Boxes 3 and 4 remain open: the existing fixture installation and separate source-Home replies supply partial evidence. Full real-provider continuity and installed authority/isolation must pass on the signed alpha.25 during 26–28 October acceptance, with Irzhy's review before stable. If that acceptance fails, the reduced version needs both commits, integration, checks and a new candidate for the 29 October go/no-go. The exception supplies no installed acceptance, review approval or signing/publication authority.

In source commit 32e552b91, the OpenRouter key check confirms the key with OpenRouter before it loads models, so Check key catches a wrong key before the first Assistant message.

Review: Codex review passed.
Release train: integrated on release/train-0.8.0-rc10 as 6e8910733. The train's Linux source gate passed on 6e8910733 (24m29s).
Mac test Home so far. This is a source Home built from c4074b463, not an installed release, and it does not have the key check fix yet.
- Anders got an Assistant reply from Venice after he approved the connection in Inbox. The Home recorded one completed Venice run, created about 25 seconds after the approval. The number of requests that reached Venice is not verified.
- A restart kept the saved OpenRouter connection. The Venice connection was added after that restart.
- A scan of 199 log and data files found no OpenRouter-style keys or Bearer headers. The scan for the Venice key has not run yet.
Still open: End and End during a reply on the current test Home. After those, the test Home is rebuilt at 6e8910733 for one OpenRouter reply with the full key and a Venice restart check. Exact CI, packaging, route-removal continuity, installed authority and isolation checks, human acceptance and the platform scope are also open.

Edited 21 Oct: the first version of this comment called the test Home an installed Home and stated the send count, restart and key scan more broadly than the evidence shows.

Mac test Home, Venice: source Home observations (21 Oct)

Build: a source Home built from c4074b463, not an installed release; Anders as admin. It does not have the key check fix (6e8910733).
Approval: OMP reports four completed Venice replies, each started within a window that Anders had approved in Inbox. Messages sent after End, or while a new request waited in Inbox, were refused (hosted_access_ended, hosted_access_pending). The number of requests that reached Venice is not verified.
End: Anders reports a refusal after End and a streaming reply that stopped when he pressed End. OMP reports two interrupted runs, which ended 33 ms and 25 ms after End (hosted_access_ended_during_reply).
Restart: OMP reports that a full restart kept the saved Venice connection and its decisions unchanged; the key file was still present with the same size.
Keys: a scan of 222 log and data files found no matches for the tested key patterns, Bearer tokens or filled key fields. It did not search for the Venice key, so that coverage is untested.
Still open:
- one successful OpenRouter reply, under the existing cutoff and fallback (decision);
- OMP owns the fix for System showing the old End state after a re-approval, and the test Home rebuild at the current candidate;
- the complete owner Inbox and connection journeys, route-removal continuity, installed authority and isolation checks on the exact candidate, exact CI and packaging, the required human review and acceptance, and the platform scope.

Edited 21 Oct: the first version called this an installed Mac proof, moved the OpenRouter reply to later acceptance, and stated approval, End, restart and the key scan more broadly than the evidence shows.

Source-Home observation, OpenRouter: one approved reply (21 Oct)

Home: the same isolated test Home, built and installed from source at c4074b463. It is not a release candidate, and it lacks 6e8910733 (OpenRouter key check) and 7823a7334 (live refresh).
Approval: Anders saved a new connection with the correct key, approved its Inbox decision and reported "openrouter confirmed working". One run started 34 s after the approval and completed, with its Runtime audit requested/completed pair. The retained Home records show one OpenRouter run in the approved window, up to the snapshot time. These are Home-side records; the upstream request count needs the provider's own logs and is not verified.
Keys: OMP reports no matches in 223 scanned files for OpenRouter-style key, Bearer-token and populated JSON-key patterns. This was not a search for the exact secret, and Venice keys have no pattern to scan for, so Venice key coverage is unverified. Excluded: the model-provider secrets, binaries, capsules, libexec and ipfs-repo.
Not observed with OpenRouter: End and End during a reply were observed with Venice (comment above). OpenRouter uses the same Runtime path; that is a source reading, not an OpenRouter observation. The restart observations stay in the earlier comment; they are not installed-candidate acceptance.
Still open for 0.8.0: Hosted AI with owner approval #85: the wrong-key check and live refresh on a rebuilt current-candidate Home, the exact CI and package proof, installed authority and isolation, route-removal continuity and the full 0.8.0: Hosted AI with owner approval #85 journeys. The review requirements are unchanged, including 0.8.0: Hosted AI with owner approval #85 (comment). The cutoff and fallback are in 0.8.0: Hosted AI with owner approval #85 (comment).

Edited by OMP on 21 Oct: scoped to the retained evidence and its stated limits.

Decision (Anders, 22 Oct): Hosted AI stays on in 0.8.0-alpha.25. This is an exception to the 22 Oct 18:00 UTC cutoff, for alpha.25 only.

Evidence so far is partial, and none of it is on a signed build:
- Saved connections and their files were kept when the candidate abddf200e was installed with the source-Home setup over a signed alpha.24 install. That run used fake keys and a local stand-in provider, so a real reply after the switch is still to be shown.
- Real OpenRouter and Venice replies, End and a wrong-key rejection were seen on source test Homes (c4074b463, 26f7c7589), which were not updated from alpha.24.
- Part of the installed authority and isolation checks passed on the abddf200e source install with fake keys. The Runtime and model-provider code on the train is unchanged since abddf200e.
0.8.0 keeps Hosted AI on only if, on the signed alpha.25 during acceptance (26–28 Oct), boxes 3 and 4 pass (saved connections after updating from alpha.24; installed authority and isolation with one real request per provider), and Irzhy's review passes.
Otherwise: the reduced version ships (00a2a87ee, c710537bf). It is not on the train yet, so it needs integration, its checks and a new candidate. It blocks new connections and sharing; existing owner connections keep working.
When: the decision is taken at the go/no-go on Thu 29 Oct (#93 schedule).

Navigation Menu

Repository navigation

0.8.0: Hosted AI with owner approval #85

Done means

29 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

0.8.0: Hosted AI with owner approval #85

Description

Done means

Activity

andersalm commented on Sep 29, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

andersalm commented on Sep 30, 2026

29 remaining items

andersalm commented on Oct 19, 2026

andersalm commented on Oct 19, 2026

andersalm commented on Oct 19, 2026

andersalm commented on Oct 19, 2026

andersalm commented on Oct 19, 2026

andersalm commented on Oct 19, 2026

andersalm commented on Oct 19, 2026

andersalm commented on Oct 19, 2026

andersalm commented on Oct 21, 2026

andersalm commented on Oct 21, 2026

andersalm commented on Oct 21, 2026

andersalm commented on Oct 21, 2026

andersalm commented on Oct 22, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions