TaskbashTop 10Deep-reviewed
by Subash Rajaseelan
What they're building
taskbash is the AI Chief of Staff for operators. It pulls every commitment you make across Gmail, Granola, calendar, and Linear into one ranked daily list, then autocompletes the work that doesn't need your brain: drafting replies in your voice, prepping you for every meeting, running follow-ups while you're heads-down. The list stays clean because the agent learns from every task you dismiss. Operators dogfooding it ship 20% more per week without staying later, and the autocomplete sharpens every week.
AI code reviewrepo: real
262-file Next.js app live at taskbash.app with Google/Granola OAuth and a real ranked-task dashboard. Genuine agent loop in lib/digest/run.ts: parallel extract (Gmail/Calendar/Granola/Linear/Slack) -> Nebius Llama-3.3-70B classifier (nebius-trace.ts called from classify/functions.ts:162) -> diff -> deliver, with full llm_calls cost tracing, Langfuse evals, and an /observability slop-rate page. All three sponsors are load-bearing and CALLED: Tavily client.search() enriches meeting attendees (enrich/tavily.ts, wired into extract/calendar.ts:127); Composio v3 @composio/core executes SLACK tools (connectors/composio.ts -> extract/slack.ts:48); Nebius via openai SDK at api.studio.nebius.ai. plus mem0 feedback loop. ⚑ Loom demo URL returned a transient 'Loom is slow' error at fetch time, but live site is strong demo evidence; repo cloned read-only, not executed.