Tasklet

Tasklet is a cloud AI-agent platform designed to take ongoing responsibility for business work rather than only answer prompts. Agents can connect to thousands of SaaS tools, direct HTTP APIs and MCP servers; use websites and a dedicated browser; run scripts and developer tools in a sandboxed command-line environment; and continue working in the cloud when the user is offline.¹²

The current product supports persistent context, shared team agents, schedules and event-driven triggers, instant apps, organization knowledge and human-approval controls for actions that should not run autonomously.¹ Users can choose models from multiple providers rather than being locked to one lab, and agents can use different model/reasoning levels for different work.²

Y Combinator independently corroborates Tasklet as an active 2026 company building a standalone cloud agent for knowledge work after the founders' earlier Shortwave product.³ Revenue, growth and investor statements on that profile remain company-supplied context rather than independent product-performance evidence.

Tasklet has a meaningful free tier: $0 per month with 300 daily bonus credits, limited usage, up to 10 executions per automation and smaller file uploads. Starter costs $25 per month for 10,000 monthly credits; Pro costs $100 per month for 40,000 credits. Paid plans also receive 600 daily bonus credits and unlock unlimited automation runs, the agent web browser and full-size file uploads.²

Custom monthly plans start at $250 and can scale substantially with usage. Tasklet also sells one-time credits at $5 per 1,600 credits with a $25 minimum; those credits remain valid for one year.² Organizations can add unlimited team members without per-seat charges and draw from one shared credit pool, which is attractive for broad adoption but makes effective cost depend on model choice, context size, tool use and automation frequency.

Tasklet's Trust & Security documentation says each agent runs in an isolated cloud environment with its own permission boundary, credentials are stored in an encrypted vault, data is encrypted at rest and in transit, and administrators can control tools and permissions per agent.⁴ The same page says Tasklet is SOC 2 compliant and has been reviewed and attested by an independent auditor; the July 9, 2026 changelog more specifically identifies the compliance milestone as SOC 2 Type I.⁵

The Privacy Policy, last updated October 3, 2025, says Tasklet may use User Content such as prompts, chats and other inputs to improve or train its own AI models unless the user opts out in settings.⁶ It separately commits to Google's Limited Use requirements for Workspace API data and says authentication tokens for integrated services are stored encrypted.⁶ Teams connecting sensitive business systems should therefore review both account-level training choices and per-agent permission scopes before deployment.

Independent product evidence is still limited. G2 has an exact Tasklet vendor/product surface tied to tasklet.ai, but currently reports zero reviews and says there are not enough reviews to provide buying insight.⁷ The descriptive G2 copy also appears less aligned with Tasklet's present cloud-agent surface, so it is not used to characterize capabilities. Zero reviews are treated as missing evidence, not a negative rating signal.

Tasklet is best suited to knowledge-work teams that need agents to move across multiple business systems and keep recurring processes running: sales operations, support, marketing, finance, research and internal operations are natural fits.¹ It is especially relevant where teams want shared agents and context, browser fallback when APIs are insufficient, and organization-wide visibility into usage and spend.

It is less suitable for buyers who require a self-hosted/open-source agent stack, a fixed unlimited-use price or a large independent review corpus before deployment. Because agents can act in connected systems, organizations should design approval gates and least-privilege tool permissions around the consequences of each workflow rather than treating model output as automatically safe.