v0.2.14

Highlights

• Multi-user shared store — ~/modl/store can be a shared symlink; store/index.yaml reconciles new users in milliseconds via modl ls
• MCP workflow tools — run_workflow (fire-and-forget), job_status, list_run_outputs exposed to AI agents
• modl status <run-id> — aggregate status of workflow runs with artifact paths
• modl outputs export <run-id> — download all outputs from a run (local or remote)
• Base64 inline images — edit: accepts data:image/...;base64,... URIs in workflow YAML
• Named image variables — images: map in workflow YAML; define character/style refs once, reference as $name
• modl edit --mask --blend — masked editing with pixel or latent (Klein) blending
• SDXL/SD1.5 offline loading — single-file checkpoints load fully offline via bundled pipeline configs
• Multiple YAML specs — modl run a.yaml b.yaml runs them sequentially
• --skip-existing/--force — resume interrupted runs without regenerating completed work
• WebP thumbnails + cache headers in the web UI
• WAL→DELETE fallback — modl remains functional even when disk is full

Dependency fixes

• diffusers pinned to >=0.38.0 (was pulling from git main)
• ai-toolkit pinned to commit 6bb8acb (was always pulling HEAD)
• Runtime staleness detection — after modl upgrade, the next command warns if Python deps need updating; modl runtime install handles the actual update

What's Changed

Full Changelog: v0.2.13...v0.2.14