CMesh v0.1.0 Linux RC2

CMesh v0.1.0 Linux RC2 is a Linux prerelease for hands-on validation of the Qwen distributed sliced execution lane.

What changed since RC1:

• Added Qwen adapter/catalog validation coverage and comparison reporting.
• Added AWS Qwen validation harnesses for 4-node distributed sliced tests.
• Fixed large stage-daemon decode responses by raising the client response cap from 1 MiB to 256 MiB.
• Added regression tests for empty, large, and panic stage-daemon decode responses.
• Validated Qwen2.5 32B real physical stage shards on 4 EC2 nodes through a single distributed decode step.

Validation status:

• Local Go test suite passed.
• Linux stable release signature smoke passed.
• Manager installer smoke passed.
• Worker installer smoke passed.
• Runtime smoke passed.
• Linux production launch gate passed.
• AWS worker test instances from the 32B validation run were terminated and cleaned up.

Known limitation:

• Qwen2.5 32B long multi-token distributed generation was progressing, but was stopped manually to control AWS cost before final completion. Do not present this RC as a completed 32B long-generation benchmark.
• Linux is the supported production test lane in this RC. macOS and Windows production installers remain follow-up work.

CMesh v0.1.0 Linux RC1

CMesh v0.1.0 Linux RC1

CMesh v0.1.0 Linux RC1 is the first public release candidate for the CMesh
Linux production path. It packages the manager, worker, verified runtime
artifact, Linux install scripts, model matrix, sliced runbook, signing metadata,
and validation docs into a signed release tarball.

What Works

• Linux manager install with install-manager-linux.sh.
• Linux worker install with install-worker.sh.
• Systemd manager and worker service path.
• Runtime auto-management for the pinned Linux amd64 llama.cpp stage runtime.
• Signed package, signed manifest, signed checksums, and signed tarball.
• Fresh-user validation from a signed tarball.
• Supported sliced model matrix for qwen2.5-14b-instruct-q4-k-m.
• Production-like sliced execution evidence across multiple Linux workers.
• Memory-aware placement evidence for a model larger than a single worker's
  declared memory allocation.
• Backup/restore, security hardening, observability, and sliced runbook docs.

Supported Scope

• Manager/worker CLI: Linux amd64 and Linux arm64.
• llama.cpp stage runtime: Linux amd64.
• Production sliced model: qwen2.5-14b-instruct-q4-k-m.
• Deployment model: private invited Linux cluster.

Not Supported Yet

• Windows production installer.
• macOS production installer.
• GPU production path.
• Trustless public worker marketplace.
• Arbitrary unvalidated Hugging Face model slicing.
• Payments, credits, reputation, or fraud resistance.

Release Assets

Upload these files together:

• v0.1.0-linux-rc.1.tar.gz
• v0.1.0-linux-rc.1.tar.gz.sha256
• v0.1.0-linux-rc.1.tar.gz.sig
• v0.1.0-linux-rc.1.tar.gz.public-key.pem

Verify

shasum -a 256 -c v0.1.0-linux-rc.1.tar.gz.sha256
openssl dgst -sha256 \
  -verify v0.1.0-linux-rc.1.tar.gz.public-key.pem \
  -signature v0.1.0-linux-rc.1.tar.gz.sig \
  v0.1.0-linux-rc.1.tar.gz

Then extract and verify package internals:

tar -xzf v0.1.0-linux-rc.1.tar.gz
cd v0.1.0-linux-rc.1
openssl dgst -sha256 -verify release-signing-public-key.pem -signature manifest.json.sig manifest.json
openssl dgst -sha256 -verify release-signing-public-key.pem -signature checksums.txt.sig checksums.txt
shasum -a 256 -c checksums.txt

Evidence

• Linux production launch gate:
  /var/folders/dp/58nxklj177sckhsl4v9p8g500000gn/T/cmesh-linux-production-launch-gate-20260620144800
• AWS beta evidence:
  /tmp/cmesh-linux-beta-deployment-20260620135153

Known Constraints

This is an RC. The supported path is intentionally narrow and Linux-first. Use
private invited workers only and do not send sensitive prompts to untrusted
workers.

v0.1.0-alpha.86

Advance model operations and capacity readiness

v0.1.0-alpha.85

Adds llama.cpp runtime cache migration. Worker runtime repair now reuses a valid legacy CMesh runtime cache before downloading a new llama.cpp release.

v0.1.0-alpha.84

Fixes stale local Worker control API token recovery. The Worker app now stops an older local control process on 127.0.0.1:9781, starts its own control process, and retries the failed action once.

v0.1.0-alpha.83

CMesh v0.1.0-alpha.83

Release candidate prepared from commit db5b854.

Highlights

• Worker runtime repair flow is now explicit in the Worker app and local control API.
• Manager separates Installed Models, Model Catalog, Model Activity, Chat, and Readiness surfaces.
• Installed model inventory now includes worker, path, size, and runtime readiness metadata.
• Model install jobs report local download progress in the Worker app.
• Worker heartbeats now include CMesh storage accounting for models, runtimes, and cache usage.
• Model install scheduling now rejects workers that have insufficient physical free disk.
• Alpha validation docs now include a release checklist and model smoke pack.

Validation Before Publishing

• Run make release-dry-run VERSION=v0.1.0-alpha.83.
• Confirm dist/release-dry-run/report.txt contains PASS: release dry-run completed.
• Confirm macOS notarization secrets are configured before publishing public DMG builds.
• After GitHub release assets are published, run:

CMESH_VERSION=v0.1.0-alpha.83 CMESH_DRY_RUN=true scripts/deploy-alpha.sh

Only deploy alpha after the asset guard passes.

Expected Assets

See expected-assets.md.

Commit Summary

db5b854 Add release candidate prep tooling
0ba09bc Harden alpha model operations
54fed5e Add worker runtime repair flow
dc29299 Separate installed model inventory
6900943 Add release dry-run script
6b109b1 Add cluster readiness dashboard
9d6b7ed Refresh model lifecycle state inline
3bea992 Add worker runtime repair controls
40188da Document runtime-ready model inventory
5c4471e Gate model chat on ready runtime
a630c0c Summarize model delete activity
d6d2649 Record model delete cleanup metadata
be5a6b1 Add worker model runtime tab
7c8622f Add scheduler debug tab
6309876 Update model workflow documentation
ad45e45 Add conversation delete controls
fb4ca7e Show worker health inventory
ce2e6b2 Improve model lifecycle operations
c6877fa Explain model scheduling eligibility
f93305d Organize model console and runtime status

v0.1.0-alpha.82

Add manual model memory editor

v0.1.0-alpha.81

Add memory controls and prompt preview

v0.1.0-alpha.80

Add model memory persistence

v0.1.0-alpha.79

Add conversation manager UI