Hand an agent a job, not your keys
Write the job in a sentence and switch on the GitHub, Linear or Slack actions it may use. Toolspoke keeps the credentials, runs it on its schedule, and logs every call.
Free to start, no card needed.
Content gapsDrafts briefs in Notion

Deal nudgerSpots deals gone quiet
Flaky testsRanks the tests that fail
PostmortemsDrafts the write-up
VulnerabilitiesLists new issues weekly
Revenue updatePosts Monday’s numbers
Every call it made, and what it posted
A morning briefing agent’s run, the way Toolspoke shows it: the time and cost, each tool call in order, and the message it left in Slack.


The jobs nobody has time for
Chores most teams put off, and the catalogue agent that does each one.
Customers pay late, and chasing them slips to next week, every week.
Overdue invoice chaser lists late Xero invoices every Monday and leaves one polite reminder draft per customer in Gmail. you press send
A red build eats the morning while three people read the same log.
CI failure explainer reads the failed run down to the step and line that broke, and says whose problem it is.
Bug reports sit in Zendesk, where engineering never sees them.
Bug report escalator turns the ticket into a Linear issue engineers can start on, without asking the customer again.
Card spend and the books never quite match at month end.
Month-end spend check matches a month of Ramp spend against the bills in QuickBooks and lists every difference with a reason. it never edits the books
Toolspoke for enterprise
Connect your internal systems, give each team its own rules, and keep credentials on your own servers when your policy asks for it.
- Your own APIs
Build a toolkit from an OpenAPI document, a Postman collection, an MCP server or a command-line tool, and use it beside Salesforce and Jira.
- Your own procedures
Write how your team does a job as a skill. Agents follow it on a schedule, when asked, or for each new item a routine finds.
- Rules per team
Each private project has its own people, toolkits and action policy. Organisation admins reach every project by role.
- Every call on record
Which agent or person called what, and the result. Credentials are encrypted with AES-256-GCM, and the agent never sees them.
Enterprise plan. Managed hosting, SSO, extended retention and an SLA. Priced by agreement.
Compare plansDescribe the job, and it builds the agent
Write one sentence. The builder proposes the instructions, the toolkits it needs, a schedule and where it should answer. You add or skip each toolkit, routine and channel, then create the agent.
- Start from a catalogue agent instead when one already does the job
- Every piece stays editable after you add it
Jobs teams hand over first
Every agent we publish does one job, and ends with a list of things it will never do.
Meeting prep brief
Each weekday morning, drafts one email briefing you on today’s external meetings from HubSpot and your past email threads.
Draft · saved, not sent · to you
Your four external meetings today
NeverSend an email. It only ever saves a draft, addressed to you.
Flaky test finder
GitHub issue · Flaky tests this week
checkout.spec9 of 40 runsauth/login.spec4 of 40 runsNeverRerun a workflow. A rerun that passes hides a flaky test.
Stale deal nudger
Slack · #sales
Acme renewal has been quiet for 14 days. Next step: ask Dana about the seat count.
NeverChange a deal’s stage, value, owner or status, or mark it won or lost.
It answers where your team already talks
Ask it in Slack, email it, mention it on a pull request, or let it wake on its schedule. It replies in the thread it was asked in.
- When someone emails it, each write waits for your go-ahead, unless you’ve allowed that action in advance
- A routine can watch for new items and wake the agent only when one matches
An example thread.
An example thread.
An example thread.
An agent never reaches past its owner
Three rules the gateway enforces on every call, whatever the prompt says.
Off means unreachable
An action you switch off can’t be called, whatever the prompt says.
The agent never holds the key
The gateway adds the credential to each call. You only ever see its last four characters.
Every call is written down
Which agent called what, and when. Take someone off a project and their agents lose it on the next call.
See what every agent spends
Tool calls are free. Each run is metered in model tokens and sandbox time, per agent, and the free plan includes $3 of that every 30 days.
Start from an agent someone already wrote
Each one does a single job you can read in full before you sign up. Add a skill to teach it a procedure, and a toolkit to reach another app.
Agents
- Inbox briefer
- AWS cost report
- Bills due planner
- Bug report escalator
- CRM hygiene report
- Cash runway monitorM
Works in the AI apps you already use
Add Toolspoke to Claude, Cursor, Codex or VS Code and sign in once. They reach the same tools as your hosted agents, behind the same switches, and every call lands in the same log.
Questions before the first agent
What is a hosted AI agent?
An agent that runs on Toolspoke with the instructions and toolkit actions you give it. It runs on a schedule, or you talk to it in Slack, email or the app.
Do I need to write code?
No. You write the job in plain language and switch on the actions it may use. Developers can also bring their own APIs, MCP servers and CLIs.
Can I use an agent I already have?
Yes. Point Claude, Cursor, Codex or any MCP client at the gateway, and it gets the same toolkits, access rules and log as a hosted agent.
What’s in the marketplace?
Agents that do one job each, skills that teach a procedure, and toolkits that connect an app. You can read all of them before you sign up.
What does it cost?
Tool calls are free. You pay for sandbox time and model tokens, and the free plan includes $3 of that every 30 days.
Can I try it for free?
Yes, without a card. Compare plans and allowances
Give one agent one job this week
Pick the job you keep putting off, and switch on only what it needs.