Repository navigation
Releases
Akasha 0.13.0-beta.3 — bêta Windows CPU et CUDA
Akasha 0.13.0-beta.3 — bêta Windows
Préversion Windows construite depuis main au commit fc08c8f98cb99f1406c12318217541aa47a9e22a (tag v0.13.0-beta.3). Elle fournit deux archives portables : une version CPU et une version CUDA pour NVIDIA. Le canal stable reste séparé.
Téléchargements
- CPU :
akasha-windows-0.13.0-beta.3-cpu.zip— SHA-2560355bf613620ec1ba682d26afb72acff6efe4a714c361810cf7c2d1db1c0d0d4 - CUDA :
akasha-windows-0.13.0-beta.3-cuda.zip— SHA-25676e23de26fd3ce124196c1f420bddf2c1f557ab316b6ea938fe3ccb32b2a9bd4
Démarrer avec START-AKASHA.cmd. Les deux paquets incluent le runtime nécessaire à leur mode d’exécution.
Vérifications de cette bêta
- CI
mainet compilation des deux paquets : réussies (CI générale, GPU, lint UI et workflow Windows, commit ci-dessus). - Contrôle d’exécution : 5/5 prompts synthétiques avec l’archive CPU et 5/5 avec CUDA ; l’archive CUDA a chargé ses bibliothèques livrées et confirmé le calcul sur GPU.
- Installation locale mise à jour depuis l’archive CUDA distribuée :
/api/statusindique0.13.0-beta.3. La migration de la mémoire chiffrée a conservé les données existantes. - Parcours Arena exécuté par Akasha avec le modèle local Ollama
qwen3.8:27b: compte agent créé, session authentifiée etGET /api/v2/mevalidé (HTTP 200). La tâche de vérification s’est terminée sans sous-agent ; aucune clé ou aucun mot de passe n’apparaît dans sa trace. Le compte de test est conservé localement avec Windows DPAPI. - Rapport de qualification joint à cette publication.
Limites connues de la qualification
- Les deux paquets ont été vérifiés sur un seul PC Windows 11 Insider Preview, équipé d’un Ryzen 7 9800X3D, 61,6 Go de RAM et d’une RTX 4080 Super 16 Go. Le paquet CPU a été forcé en mode CPU sur ce PC ; aucun second PC physique sans GPU n’a été testé.
- Les essais de 5 prompts par paquet sont des contrôles de fumée. Ils ne constituent pas la qualification complète des modèles à 95 % sur trois passages.
- Cette publication ne revendique pas une qualification exhaustive de l’OCR sur des scans PDF ou des photos.
Cette version est une bêta de test ; les limites ci-dessus font partie du rapport de qualification.
Akasha 0.13.0-beta.1 — Windows CPU et NVIDIA CUDA
Préversion technique Windows d’Akasha, construite depuis main à la révision a93e938485fa32ce8645ed6a8b1394bc851b2acf.
Télécharger et essayer
- CPU :
akasha-windows-0.13.0-beta.1-cpu.zip. - NVIDIA CUDA :
akasha-windows-0.13.0-beta.1-cuda.zip; les bibliothèques CUDA nécessaires sont incluses, un pilote NVIDIA compatible reste requis.
Décompressez le paquet complet dans un nouveau dossier, puis double-cliquez sur START-AKASHA.cmd. Gardez votre installation habituelle séparée. Le paquet utilise son propre profil bêta et le port local 3877. Suivez les trois étapes de préparation puis essayez le dossier fictif : le montant attendu est 42 €, avec une échéance au 10 novembre 2026. Ouvrez les références pour vérifier l’original.
Les poids des modèles ne sont pas inclus. Les essais utilisent des modèles locaux réellement téléchargés et identifiés dans le rapport. Windows PowerShell 5.1 et WebView2 sont nécessaires.
Vérifications publiées
Le rapport qualification-windows-0.13.0-beta.1.md détaille les essais sur les exécutables extraits des deux ZIP : documents et OCR sourcés, notes en actions et rappels confirmés, veille et reprise après arrêt, sauvegarde chiffrée et restauration isolée. Les reçus, réponses, empreintes, observations de l’interface et scripts figurent dans preuves-windows-0.13.0-beta.1.zip. SHA256SUMS.txt permet de contrôler les téléchargements.
Les essais préparatoires ont conduit à corriger les questions interprétées comme un prénom, le défilement de l’accueil documentaire et la provenance des paquets Windows. Les exécutables publiés sont reconstruits après ces corrections.
Limites de cette bêta
Le pilote avec données personnelles n’est pas ouvert. Cette préversion permet des essais avec les documents fictifs fournis. La mémoire long terme et les sauvegardes sont chiffrées ; les conversations, notes et documents importés restent hors du chiffrement applicatif.
Les deux modes ont été éprouvés sur un poste Windows 11 équipé d’une RTX 4080 SUPER. Une machine physique sans GPU, un Windows vierge, la restauration sous un autre compte Windows, la mise à jour depuis une ancienne version et les trois passages réels du corpus complet restent à qualifier. Aucun modèle ni profil matériel n’obtient le label « recommandé » sur la base de ces essais.
Cette préversion ne remplace pas la version stable proposée par les mises à jour automatiques.
Défauts connus de l’export PDF : contraste et libellés de limites à améliorer ; la fenêtre d’impression peut se réafficher après l’enregistrement. Annulez cette fenêtre supplémentaire : le fichier enregistré reste disponible. Les observations sont conservées dans le rapport.
v0.12.1
What's Changed
Other
- fix(release): restore Linux WebKit subprocess permissions by @azerothl in #261
- fix(release): verify Linux desktop rendering in smoke test by @azerothl in #262
- fix(release): repair Linux UI smoke quoting by @azerothl in #264
- build(tauri): upgrade CLI for actionable DMG errors by @azerothl in #265
- Add Akasha application module hub by @azerothl in #267
- Fix Ollama context discovery and chunk oversized sources by @azerothl in #268
- Fix stale human-input dialogs and silent reply failures by @azerothl in #269
- feat(update): mises à jour Windows sélectives avec confirmation by @azerothl in #266
- Add strict free-model routing with automatic catalog refresh by @azerothl in #270
- chore(release): prepare Akasha v0.12.1 by @azerothl in #271
Full Changelog: v0.12.0...v0.12.1
v0.12.0
What's Changed
Other
- feat(studio): park swarm worktrees until explicit integrate by @azerothl in #138
- Corriger les neuf constats UI et accessibilité by @azerothl in #189
- Corriger les quatre points restants de l’audit UI by @azerothl in #224
- fix(studio): fiabiliser les résultats et contraindre les rôles P0 by @azerothl in #162
- feat(studio): unifier les capacités et les workflows de skills du projet by @azerothl in #164
- feat(studio): task-scoped browser PTY and MCP observation by @azerothl in #166
- fix(research): préserver les preuves et corriger les sujets Unicode by @azerothl in #225
- fix(research): preserve source evidence and confirmed cancellation by @azerothl in #233
- feat(studio): versioned validated tool contracts and native calls by @azerothl in #167
- feat(studio): durable checkpoints and explicit incident recovery by @azerothl in #168
- feat(studio): LSP diagnostics navigation and reviewed rename plans by @azerothl in #169
- fix(research): preserve source bounds and validate report claims by @azerothl in #234
- docs: guide utilisateur Companion LAN (pairing, discovery) by @azerothl in #171
- docs: COMPANION_SPEC.md — protocole Companion LAN (daemon) by @azerothl in #175
- Mémoire en clair : test memory.db + doctor honnête (#190) by @azerothl in #192
- doctor: Ollama/Rbitnet optionnels si embarqué OK ; spec YAML hors erreur release by @azerothl in #201
- fix(env): parse export prefix in akasha.env / connectors.env by @azerothl in #177
- fix(companion): alimenter notify depuis NotificationStore by @azerothl in #178
- fix(ui,cli): textes dossier de données par défaut (Fixes #194) by @azerothl in #195
- Fixes #198 by @azerothl in #199
- fix(daemon): bump pdf-extract for patched lopdf (Fixes #203) by @azerothl in #206
- test(akasha-vault): minimal vault tests by @azerothl in #213
- test(release): smoke premier démarrage Linux CPU v0.11.0 (#141) by @azerothl in #196
- docs: reconcile feature registry and v0.11.0 closure by @azerothl in #161
- fix(studio): worktree path boundaries and registry hardening (#145) by @azerothl in #146
- Progress on #143 by @azerothl in #200
- docs: dossier de données par défaut et titres Markdown (Fixes #191) by @azerothl in #193
- feat(studio): explicit Git GitHub PR and CI delivery by @azerothl in #170
- feat(studio): modeles par tache et objectifs mecaniques persistants by @azerothl in #237
- [CI] Réduire les recompilations et cibler les vérifications by @azerothl in #165
- Progress on #141-#143 and #147-#150: release, CUDA and CI validation by @azerothl in #239
- fix(release): align Tauri JS and Rust versions by @azerothl in #240
- fix(studio): droits de projet utilisables avec une configuration partielle by @azerothl in #238
- fix(release): validate staged and packaged release archives by @azerothl in #241
- fix(doctor): treat optional release runtimes as informational by @azerothl in #242
- ci(ui): add ESLint and path-filtered UI lint workflow (Fixes #220) by @azerothl in #221
- Fixes #209 — correct broken relative links in spec docs by @azerothl in #210
- Progress on #217: rustc warnings outside open PR overlap by @azerothl in #218
- Progress on #222: fix(ui) ESLint errors outside open-PR files by @azerothl in #223
- fix(daemon): unify localized PDF extraction failure messages (#211) by @azerothl in #216
- Fixes #207 by @azerothl in #208
- docs(companion-lan): align notify snapshot with NotificationStore (#173) by @azerothl in #202
- fix(vault): corrupted vault.enc returns explicit error (fixes #214) by @azerothl in #215
- fix(ui): clear remaining lint errors and enforce lint by @azerothl in #244
- chore(deps): update compatible audited crates by @azerothl in #245
- fix(research): validate citations and explicit source URLs by @azerothl in #243
- docs(security): clarify plaintext memory storage by @azerothl in #246
- test(research): valider le rejet des citations non consultées by @azerothl in #247
- [Release] Documenter la validation locale CUDA v0.11.0 by @azerothl in #249
- [Qualité] Supprimer les warnings rustc restants by @azerothl in #248
- fix(companion): protect routes and pairing by default by @azerothl in #250
- fix(plugins): upgrade Wasmtime sandbox runtime by @azerothl in #251
- feat(evals): add real agentic acceptance runner by @azerothl in #253
- chore(deps): modernize HTTP, NATS and LRU stacks by @azerothl in #252
- [docs] Document v0.11.0 first-boot validation gaps by @azerothl in #254
- chore(release): préparer v0.12.0 by @azerothl in #255
- fix(evals): enforce policy on virtual workspace writes by @azerothl in #257
- feat(studio): planifier et dispatcher le travail depuis le Kanban by @azerothl in #256
- fix(studio): prevent repeated context window failures by @azerothl in #258
- release: prepare Akasha v0.12.0 by @azerothl in #260
Full Changelog: v0.11.0...v0.12.0
v0.11.0
What's Changed
Other
- chore: Apache-2.0 license (LICENSE + README) by @azerothl in #127
- docs: roadmap v0.11.0 (proposition post-v0.10) by @azerothl in #128
- docs: interactive architecture map + regenerate on each release by @azerothl in #129
- P0/P7: message FR embarqué indisponible + sync docs + dette Partially documented by @azerothl in #131
- docs(P9): interop skills/plugins daemon ↔ skills/modules akasha-os by @azerothl in #130
- feat(p1): GET /api/capabilities + AR bench scaffolding (v0.11) by @azerothl in #132
- refactor(daemon): P3 api.rs extract — tool dispatch + remaining routes by @azerothl in #133
- feat(mcp): P4 must — MCP Implemented (CLI, policy gate, docs) by @azerothl in #135
- P5/P6-B3/B4: cron watch wakeups + subagent inspect/cancel-tree by @azerothl in #134
- refactor(daemon): P3 extract run_message_via_llm — api.rs <8k by @azerothl in #136
- release: v0.11.0 closeout — docs, architecture, version bump by @azerothl in #137
Full Changelog: v0.10.0...v0.11.0
v0.10.0
What's Changed
Other
- feat(companion): LAN daemon APIs for the ESP32 client by @azerothl in #122
- feat(llm): compact prompts and 8k context for mid-size Ollama by @azerothl in #123
- fix(ci): EmbeddedLlm::preload + InfoTip for llama-cpp / ui-e2e by @azerothl in #124
- release: merge local-inference (v0.10.0) into main by @azerothl in #125
Full Changelog: v0.9.0...v0.10.0
v0.9.0
What's Changed
Features
Other
- feat(code-studio): modes studio, planner read-only, meta résumé/polit… by @azerothl in #106
- Design studio by @azerothl in #108
- Code studio by @azerothl in #105
- Clean code by @azerothl in #112
- Update documentation structure and content for user and developer guides by @azerothl in #111
- Evol agnt 2 by @azerothl in #110
- feat: enhance task event handling with schema versioning and improved… by @azerothl in #113
- feat(studio): add task-scoped git worktree orchestration by @azerothl in #115
- Aipm by @azerothl in #114
- Evol agnt next by @azerothl in #116
- Evol agnt by @azerothl in #109
- Caldav by @azerothl in #118
- Embedded model by @azerothl in #119
- 0.9.0 by @azerothl in #117
- Fix Windows CUDA release build by restoring explicit
cudartlinkage by @azerothl with @Copilot in #120
Full Changelog: v0.8.0...v0.9.0
v0.8.0
What's Changed
Other
- Fix
akasha doctor --fixbehavior for prebuilt zip installs by @azerothl in #93 - Cursor/akasha doctor fix failures 04fe by @azerothl in #94
- fix: contract fenced-JSON stripping and clippy PI literal by @azerothl in #95
- Mode autonome mission by @azerothl in #97
- Graph workspace by @azerothl in #98
- 0.8.0 by @azerothl in #96
- Revert "Graph workspace" by @azerothl in #99
- Graph workspace by @azerothl in #100
- Merge pull request #100 from azerothl/graph-workspace by @azerothl in #101
- Axi by @azerothl in #102
- ci: fix "No space left on device" failures on GitHub Actions runners by @Copilot in #103
Full Changelog: v0.7.0...v0.8.0
v0.7.0
What's Changed
Other
- [WIP] Add new orchestrator for LLM router category by @Copilot in #65
- Fix debug telemetry leak, dead code, fenced-block tool injection, and path traversal in deliverables by @Copilot in #66
- Improve agent pipeline: direct fast-path for conversational queries, sub-agent self-evaluation, and per-step deliverable retry by @Copilot in #67
- fix(orchestrator): carry ORCH_DISK_DELIVERABLES_MARKER on deliverable-retry messages by @Copilot in #68
- fix: don't advertise policy-blocked file tools in orchestrated deliverable prompts by @Copilot in #69
- fix(orchestrator): require keyword signal alongside length for project-like classification by @Copilot in #70
- fix(orchestrator): preserve tail of long briefs in retry decomposition prompt by @Copilot in #71
- fix: don't satisfy PDF/PPTX deliverables with plain-text stubs by @Copilot in #72
- fix: resolve
workspace:/paths inedit_file,search_replace,apply_patch, andfile_difftool handlers by @Copilot in #73 - fix(orchestrator): drop absolute synthesis length threshold by @Copilot in #74
- fix: skip sloppy tool-prefix normalization inside multiline tool bodies by @Copilot in #75
- fix(orchestrator): populate deliverables in deterministic fallback plan from message workspace paths by @Copilot in #76
- fix: use parse_tool_calls() for sloppy-prefix detection in stream/response merge by @Copilot in #77
- fix(orchestrator): preserve plan step metadata when code→conversation override fires by @Copilot in #78
- fix: recognize markdown-style tool headers as multiline-body terminators by @Copilot in #79
- fix: apply lineage-root plan UUID rewrite to partial-edit workspace tools by @Copilot in #80
- Remove bare
agendakeyword from direct-mode classifier by @Copilot in #81 - fix: preserve absolute deliverable paths instead of stripping leading
/by @Copilot in #82 - New orchestrator by @azerothl in #64
- Fix review feedback: security, correctness, and completeness issues across 0.7.0 by @Copilot in #84
- Fix review feedback: memory leak, debug logging, TaskStore/await, env safety, workspace hardening by @Copilot in #86
- Fix TaskStore held across
.awaitin MainAgent::handle_message by @Copilot in #87 - Global enhancement by @azerothl in #85
- fix(policy): close path traversal bypass in can_read/can_write by @Copilot in #89
- New orchestrator by @azerothl in #88
- Fix agent routing: fast-path weather/news queries, strip thinking tokens, propagate target_agent by @Copilot in #90
- Fix agent hallu by @azerothl in #91
- Add AGENTS.md with Cursor Cloud development environment setup by @azerothl in #92
- 0.7.0 by @azerothl in #83
Full Changelog: v0.6.0...v0.7.0
v0.6.0
What's Changed
Other
- fix(gateway): pass MessageEnvelope by value to handle_envelope in all callers by @Copilot in #50
- feat(gateway): Phase 1 AI OS - Gateway layer with MessageEnvelope and… by @azerothl in #49
- fix(task-os): correct TaskResumed event emission and route matching by @Copilot in #52
- feat(taskOS): Ai os/phase 2 task os by @azerothl in #51
- fix(policy-engine): deny-by-default, restrict Running pause, nil UUID in gateway, state transition tests by @Copilot in #54
- feat(policy-engine): Ai os/phase 3 policy engine by @azerothl in #53
- feat: Ai os/phase 4 resilience by @azerothl in #55
- feat(observability): Phase 5 AI OS - Timeline and metrics endpoints by @azerothl in #56
- feat(cockpit): Phase 6 AI OS - GET /api/agents and GET /api/plugins by @azerothl in #57
- feat(calendar): Phase 7 AI OS - Schedule exceptions API by @azerothl in #58
- fix: address 7 PR review issues in gateway/task-lifecycle layer by @Copilot in #60
- Fix XSS in MarkdownContent, stt whitespace validation, speech_transcribe description, timeline comment, lockfile version by @Copilot in #61
- Fix review comments: resume_task race, audio format mismatch, exception endpoint security, docker fallback by @Copilot in #62
- Fix: orchestrator TaskStore scoping, browser session timeout, timeline ordering, TTS voice ref leak by @Copilot in #63
- 0.6.0 by @azerothl in #59
Full Changelog: v0.5.0...v0.6.0