Changelog

Oct 2

• Expanded registrar support — The provisioner now supports domains at GoDaddy and Namecheap, not just Spaceship. Subdomain runs no longer require a registrar credential.
• Built with LiveGraph in site footer — Every bootstrapped app's scaffold now includes a 'Built with LiveGraph' credit and link in the site footer.
• Automation Instances build-status page — Instantiate now creates a durable record for each automation instance, with a new build-status page that composes the latest build run's hop timeline, key markers from bootstrap nodes, loop queue snapshots, parked awaiting_approval runs, and a needs-you list.
• Streamlined Automation Wizard — The automation wizard's done step now renders 'Start the build' and named graph links. Advanced operator knobs are collapsed, and domain/githubOwner fields offer datalist suggestions.
• Improved fallback for 'request too large' — Groq's 'request too large' (413 TPM) refusal is now classified as a 'model' class failure, so the request will try the next fallback target with a bigger ceiling.
• Enhanced Bootstrap with self-migration and secrets — Scaffold now includes instrumentation.ts to run drizzle migrate() at boot. The provisioner reads .env.example's required secrets and generated values. Deploy-verify probes / to catch hollow-health deploys.
• Web: trim header nav, add /mcp link — 'Runs' and 'Approvals' have been removed from the header navigation; their functionality is covered by the /overview page. 'Usage' moved to the account menu. A new /mcp link has been added to the header.
• Web: add /mcp connector catalog page — A curated catalog of MCP servers is now available at /mcp, with server details, configuration actions, and add-to-node wiring flows for signed-in users.
• Credential test endpoint with per-provider probe registry — A new endpoint POST /api/user-credentials/:provider/test has been added, allowing credentials to be tested with provider-specific probes.
• Add Replace affordance for MCP server keys — The MCP server key rows now include a 'Replace' option for easier credential management.
• Surface credential setup hints in AutomationWizard preflight step — The AutomationWizard now provides hints for credential setup in the preflight step.

Sep 29

• Admin user + config management — The admin surface gets tabs: /admin/users is a searchable account table with per-user detail, sign-out-everywhere, plan changes, and daily budget caps; /admin/config shows which features and allowlists this deployment actually has. Analytics gains an active-users chart and grouped error signatures.
• Per-tool approval gates — A node can now pause a run mid-hop when the agent requests a risky tool call — reads stay automatic, while write/delete calls park in /approvals with the exact arguments shown before they execute.
• Connected accounts + new automation packs — Connect a Google account once from Settings and agents get read-only Search Console, Analytics, and Business Profile access via short-lived tokens. New packs: Google Site Health and Cloudflare Ops — scheduled reports that diff each day's snapshot against the last run.
• MCP marketplace picker — Browse PulseMCP's MCP catalog from a node's MCP settings; results show whether each server is covered by your operator allowlist before you pick it.
• Standalone analytics dashboard — Nashville's /admin/analytics gets a unified range picker driving traffic and operational stats together — signups, runs, failures, spend — plus funnel breakdowns and linked error rows.
• Quieter, dismissable alerts — Needs-you cards (cost spikes, weak routing) can be dismissed per-hop and stay dismissed across sessions; cost-spike flags now ignore sub-cent noise, and graph cards get an expandable response preview.

Sep 23

• Site footer + legal pages — Real footer on public pages with Product/Support/Legal columns, "Site by Canweb Ltd." attribution, and new /privacy and /terms pages.
• Community pages — /docs, /status (live health), and this changelog.
• Cancel from anywhere — Cancel a running/parked run from the run page, the runs list, and the overview feed.

Sep 22

• Results surface on the overview — Every recent run shows its question and the answer your agents produced, failures carry their error inline, and pending approvals get a "Needs your decision" queue with inline Approve/Reject.
• Rerun — Retry any finished run on the current graph config; provenance links back to the source run.
• Queue watchdog — The worker now detects a wedged queue connection, self-heals, re-reconciles schedules after a Redis bounce, and exits loudly if it can't recover. Redis + worker lifecycle events are logged.
• MCP observability — Granted tools and each tool call are logged by slug; silent "zero tools granted" hops are now distinguishable.
• More providers — Groq, Cerebras, Mistral, GitHub Models, and NVIDIA join the OpenAI-compatible host table; stored credentials can carry a custom baseUrl.
• Schedule hygiene — Triggers auto-disable after repeated no-entry skips instead of burning budget.

Sep 20

• An easier /overview — The multi-project dashboard: every graph's health, pending approvals, trigger status, and recent activity in one place.
• Stall rotation — A per-attempt request timeout moves a hop to its next fallback target early instead of burning the whole hop deadline on a stuck model.
• A better landing page — The hero now plays the mid-run reroute live instead of a frozen frame.