Tasklet review, pricing and verdict

Best for operations, sales and finance teams that want recurring back-office jobs handled by shared agents instead of one AI subscription per person. It replaces hand-built workflow chains where the job needs judgment. Its changelog showed a steady run of releases in October 2026, but no independent benchmark exists.

Tasklet is a business AI agent platform from a San Francisco company with a team of 12 as of October 2026. Agents work in a private cloud workspace with a filesystem, database and browser, connect to thousands of apps, MCP servers and APIs, and run on a schedule or when an event fires, even while nobody is logged in.

Maker: Tasklet · Autonomy: semi autonomous · Maturity: GA

Underlying models: GPT-6.1 Sol, Claude Sonnet 5.5, Claude Haiku 5.5, Gemini 4 Argon

About Tasklet

Tasklet is an AI agent platform for business teams, made by a San Francisco company whose CEO, Andrew Lee, says he founded Firebase and sold it to Google (Tasklet careers page, October 2026). You describe a job in plain English, connect your tools, and the agent does the work in the cloud, either on a schedule or when something happens, such as a new email or a new support ticket. That trigger loop is what separates it from a chat assistant: the agent keeps working after you close the tab, and it asks for approval before it uses a tool you have not enabled.

Each agent gets its own cloud computer, filesystem, database and browser, according to Tasklet's guide. It connects to thousands of apps, any MCP server (a standard way for AI tools to plug into software) and any HTTP API, and it can drive a browser for sites that have no API. Threads under one agent share its memory, knowledge and connections. You choose the model per thread, and the picker includes GPT-6.1 Sol from OpenAI, Claude Sonnet 5.5 from Anthropic, and Haiku and Gemini options. Auto mode, added in October 2026, picks a model for each task.

The examples Tasklet shows are back-office jobs: enriching new leads and logging them in a CRM, matching Stripe payouts to QuickBooks invoices, and mirroring Zendesk tickets into Linear. Those are vendor demos and customer stories published by Tasklet, not independent tests. The fit is a team that wants one shared set of agents across sales, support, finance and admin, rather than a separate AI subscription for every person. It is a poor fit where a wrong action is costly and nobody reviews the output, and where a task hits a login that needs an SMS code or a CAPTCHA, because Tasklet's guide says you may have to take over the browser for those.

Pricing is credit-based and billed to the organization, so the cost answer below has the numbers. Tasklet's careers page names ChatGPT Work and Claude Cowork as its rivals, and says its small team competes by working with models from any lab. On HokAI, the nearest like-for-like agent is ChatGPT Work, the text-first personal option is Poke, and an open-source agent you host yourself is OpenClaw. Other always-on agents on HokAI include OpenAI dots, Gemini Spark and the general-purpose agent Manus. If your jobs are fixed if-this-then-that chains with no judgment, a workflow tool such as Zapier or Lindy may be simpler. The changelog lists 11 dated updates between 2 and 20 October 2026, including per-agent email addresses and saved browser logins.

Pricing

Pricing read from tasklet.ai/pricing on 22 October 2026. Free: a first agent on trial credits, no card needed. Paid monthly tiers: Starter at $25 (10,000 credits), Pro at $100 (40,000), and a Custom slider from $25 up to $10,000 where $250 gives 100,000. Paid tiers include 600 bonus credits per active day, unlimited automation runs and the dedicated browser. Top-ups: $5 buys 1,600 credits, valid a year. Unused monthly credits are lost at renewal. Billing is by organization with no per-seat fee; enterprise terms are quoted.

Key Features

• Schedules and event triggers: Runs automations at set times or when an event occurs, such as a new Gmail message or a new Zendesk ticket, and keeps going in the cloud while you are offline (Tasklet homepage).
• Own cloud computer per agent: Each agent has a sandboxed command line, filesystem, database and browser, so it can run scripts, process files and sign in to sites that have no API (Tasklet guide and careers page).
• Thousands of connections plus MCP: Connects built-in apps such as Gmail, Slack, HubSpot, Salesforce, Stripe and QuickBooks, any MCP server and any HTTP API, with tool permissions set separately for each agent.
• Model choice and Auto mode: You pick a model and reasoning level per thread from Anthropic, OpenAI, Google and open-source labs, and Auto mode, released in October 2026, selects one for each task.
• Shared agents, connections and memory: A team shares agents, connections, files and memory in one organization, and admins can view usage and cap spend for individual agents (Tasklet pricing FAQ).
• Instant Apps: Turns a prompt into a dashboard, tracker, calculator or approval workflow tied to your connected data, which can be installed as its own Mac app (Tasklet guide).
• Email and text channels: A dedicated Tasklet email address can be given to any agent (added 14 October 2026), and threads can also be reached by iMessage and RCS text message (Tasklet changelog and guide).

Strengths

• Billing is per organization with a shared credit pool, so adding a colleague does not raise the bill, which suits teams that would otherwise buy one AI seat per person (Tasklet pricing page).
• Tasklet says an independent auditor attested it as SOC 2 compliant, that credentials sit in an encrypted vault and that customer data is not used to train third-party models (Tasklet Trust & Security page).
• Not locked to one lab: on 22 October 2026 the model picker spanned OpenAI, Anthropic and Google releases, and Auto mode routes work between them.
• Ships quickly: its changelog shows 11 dated entries between 2 and 20 October 2026, from parallel subagents to saved browser logins with 1Password.

Weaknesses

• Cost is metered in credits that depend on task complexity, context size, active tools, automation frequency and chosen model, and monthly credits do not roll over, so a busy automation is hard to budget in advance (Tasklet pricing FAQ).
• Evidence is vendor-published: the only customer stories are two Tasklet case studies, and no independent benchmark or review was found, so the benchmark fields stay empty.
• Some logins still need you: SMS codes, push approvals, passkeys and CAPTCHAs may require you to take over the browser preview (Tasklet guide).
• A team of 12 (Tasklet careers page) is up against agents from OpenAI and Anthropic that Tasklet itself names as rivals, so long-term support and pricing stability are open questions.

Visit Tasklet Official Site