v0.4.0 — Knowledge Autonomy

AIMA v0.4.0 — Knowledge Autonomy

v0.4.0 closes the Edge↔Central knowledge loop: Explorer can discover work, benchmark and tune models, harvest evidence, sync upstream, and consume Central advisories and scenarios with a stable lifecycle.

Highlights

• Explorer Agent Planner replaces the old single-shot planner with a document-driven PDCA workflow.
• Explorer Web UI MVP adds live Overview and Runs surfaces backed by read-only inspector MCP tools.
• Central Advisor / Analyzer and Sync v2 deliver advisories, scenarios, and feedback across the Edge↔Central path.
• aima-service device registry integration gives edge devices a unified cloud identity.
• Onboarding cold-start wizard, multi-modal benchmark support, and MCP/tooling consolidation all land in this release.

Assets

This release includes:

• Raw binaries for macOS arm64, Linux amd64, Linux arm64, and Windows amd64
• macOS app bundle zip
• Linux desktop bundles for amd64 and arm64
• checksums.txt for integrity verification

Known Issues

• Onboarding smoke coverage is still missing aibook, m1000, and metax-n260 refreshes from this release window.
• Multi-modal benchmark evidence is still incomplete until w7900d and aibook are both fully available for the remaining coverage.
• The post-tag smoke matrix passed on all 7 reachable hosts, but aibook, m1000, and metax-n260 remained unreachable, so U13 stays as a known coverage gap rather than a product failure.
• Central SQLite/Postgres contract closure is complete, but the final cross-repo regression evidence lives in aima-central-knowledge.
• Overlay YAML hardware identity matching remains intentionally coarse in v0.4 and is deferred to v0.5.

Release v0.3.3

Full Changelog: v0.3.2...v0.3.3

Release v0.3.2

What's Changed

• hotfix(v0.3.2): simplify onboarding manifest generation by @rjckkkkk in #35

Full Changelog: v0.3.0...v0.3.2

Release v0.3.0

What's Changed

• feat(engine): download progress, auto-pull, and aima run by @skyguan92 in #31
• fix(catalog): add glm-asr-nano huggingface source by @rjckkkkk in #29
• docs(openclaw): clarify aima-tts skill contract by @rjckkkkk in #30
• feat(ui): add onboarding drawer for new users by @rjckkkkk in #32

New Contributors

• @rjckkkkk made their first contribution in #29

Full Changelog: v0.2.0...v0.3.0

Stack Bundle 2026-02-26

Non-product bundle tag replacing legacy v0.1.0-images for stack/airgap assets. Product releases remain vX.Y.Z only.

v0.0.1 — Initial Release

AIMA v0.0.1

First official release of AIMA (AI-Inference-Managed-by-AI) — a Go binary that manages AI inference on edge devices.

Highlights

• 80 MCP tools for AI Agents to operate inference workloads
• Hardware detection: NVIDIA, AMD, Intel, Huawei Ascend, Moore Threads, MetaX, Hygon DCU
• 3 runtimes: K3S (Pod), Docker (container), Native (exec)
• YAML knowledge base: 8 hardware profiles, 18 engine configs, 13 model configs
• Knowledge resolver with L0-L3 config layering
• Pod YAML generator with dynamic GPU resource names
• SQLite state store (zero CGO, modernc.org/sqlite)
• Embedded Web UI (Alpine.js SPA)
• Terminal dashboard (Bubble Tea TUI)
• Central knowledge server for fleet sync
• Offline-first: all core functions work with zero network

Pre-built Binaries

File	OS	Arch	Target Devices
aima-darwin-arm64	macOS	arm64	Apple Silicon Macs
aima-linux-arm64	Linux	arm64	NVIDIA GB10, Moore Threads M1000, Ascend 910B
aima-linux-amd64	Linux	amd64	x86_64 servers (RTX 4090, Hygon DCU, MetaX N260, AMD)
aima-windows-amd64.exe	Windows	amd64	Windows workstations with NVIDIA GPU

Usage

# Download and make executable
chmod +x aima-linux-amd64

# Quick start
./aima version
./aima hal detect
./aima engine list
./aima model list

SHA256 Checksums

14c037eba22c9272cc9d6f449ba66b902328d0eac7a6d61f82a3f492a21c641a  aima-darwin-arm64
ca9024cb5f343e5847a2dd8fc2d6f6b05c16b2509ecd6e7ffc9b7873856aa12e  aima-linux-amd64
650ef67a6df0568d11ace8938149de3f04ba45c703bc1466dff5224f909533bb  aima-linux-arm64
4bbf8c40234933c29be7a9545a37b1e67f4a87379fd3268d802fe15ce52b9aeb  aima-windows-amd64.exe

Full Changelog: https://github.com/Approaching-AI/AIMA/commits/v0.0.1

v0.1.0 Airgap Images (Deprecated)

Deprecated: this legacy bundle release is retained for compatibility only.
Use bundle/stack/2026-02-26 for stack and airgap assets going forward.
Product releases remain vX.Y.Z only.

---

Pre-built container image tars for airgap/offline installation.

HAMi v2.4.1 Airgap Images

Contains projecthami/hami:v2.4.1 + liangjw/kube-webhook-certgen:v1.1.1, exported via docker save.

File	Arch	Size
hami-airgap-images-arm64.tar	linux/arm64	237 MB
hami-airgap-images-amd64.tar	linux/amd64	398 MB

Usage

These files are automatically downloaded by aima init when airgap mode is configured.
To manually place them:

mkdir -p ~/.aima/dist/linux-$(uname -m | sed 's/aarch64/arm64/;s/x86_64/amd64/')
cp hami-airgap-images-*.tar ~/.aima/dist/linux-*/hami-airgap-images.tar