Hire AI staff that do the work.
Helm AI agents are autonomous staff that execute professional services work: running projects, managing accounts, chasing invoices, qualifying leads, and drafting client deliverables. Each agent has a defined role, scoped tool access, and an autonomy level you control. Every agent operates on the same records as your human team.
A role, a toolkit, a level of autonomy.
Three decisions build an agent. Customize further, or deploy the defaults and refine over time.
Step 1
Pick a role
Start from a pre-built role like Project Manager or Account Manager. Or define a custom role with your own name, color, and defaults.
Step 2
Scope the tools
Enable the tool groups the agent needs. Toggle read or read-write per group. Add 1000+ external integrations to extend reach.
Step 3
Set the autonomy
Observe (watch), Suggest (propose + approve), or Auto (execute). Override per channel so email can still require approval.
Reach your agents where you work.
Chat, email, Slack, or on a schedule. Each channel has its own permissioning and autonomy override.
In-app chat
Open the chat panel in Helm and talk to any agent. Page context comes along for free.
Reach agents on any mailbox the workspace connects. Replies route back through the agent.
Slack
@mention an agent in any channel it's invited to, or DM the workspace DM handler.
Scheduled runs
Run agents on a daily, weekly, or monthly cadence. Output to Slack, notifications, or a task comment.
Agents that report and delegate.
Agents don't work in isolation. Give each one a place in your org chart, then let them hand work to each other the way a real team would.
Agents fill org-chart roles
Put an agent in a real seat on the org chart, with a title, responsibilities, and a department. The agent carries that identity into every task it works.
Reporting lines
Roles report to roles, so the chart shows who answers to whom. People and AI agents sit on the same structure, in the same departments.
Delegate to a teammate
An agent can hand a task to a peer agent by name, with handoff instructions. The specialist picks it up and starts working automatically.
Specialize and chain
A research agent gathers, then hands off to a copywriter agent to draft. Each agent does the part it does best, without you relaying between them.
Pre-built agents ready to install.
Deploy a template on day one. Customize the instructions, tools, skills, and autonomy as you learn what your team needs.
Klaviyo Revenue Agent
byHelmReviews your Klaviyo flows, campaigns, and deliverability weekly, then ranks what to fix.
WordPress Content Agent
byHelmDrafts and publishes WordPress content, manages media and comments, and watches site health.
Payment Recovery Agent
byHelmRecovers failed payments with personalized three-touch outreach, and escalates the high-value accounts.
Account Manager Agent
byHelmRuns client check-ins, drafts recaps and QBRs, tracks account health, and flags at-risk accounts before renewal.
Content Strategist Agent
byHelmPlans editorial calendars, maintains brand voice, and drafts posts, case studies, and client updates from real project data.
Operations Manager Agent
byHelmWatches capacity, profitability, and AR, surfaces bottlenecks and at-risk projects, and runs operational summaries automatically.
Build a completely custom agent
Your role, your instructions, your tool scope. Skip the templates when your workflow is too specific for an off-the-shelf agent.
Browse skill templates
Hundreds of predefined skills plus custom skills. Plug domain expertise into any agent, pre-built or custom.
Tune every dimension of how agents work.
The full tuning surface, shared across every role. Start simple, go deep as you scale.
Pre-built or custom roles
Start from a template. Or build a custom role with its own default instructions, tools, and autonomy.
Scoped tool access
Enable specific tool groups (tasks, projects, billing, calendar, analytics, memory, and more). Toggle read or read-write per group.
Three autonomy levels
Observe (watch silently), Suggest (draft and wait), or Auto (execute directly). Flip levels as trust builds.
Per-channel autonomy override
A fully autonomous agent can still require approval on email. Rare in AI tools, essential for client-facing work.
Custom instructions and skills
Layer your own instructions on top of the role's defaults. Enable skill modules for domain expertise and house style.
Memory
Agents remember past decisions, scoped to workspace, account, or project. Searchable. Optional. Disable per agent.
Schedules
Run agents on a daily, weekly, or monthly cadence. Output to Slack, in-app notifications, or a task comment.
Spending limits
Set a monthly USD cost cap per agent. Runs pause when the budget is hit. Usage tracked per run and per month.
Full audit trail
Every agent run is logged with the tools called, inputs, outputs, tokens, and cost. Browse history per agent.
Already running your own agents? Give them a business to run.
Helm has a CLI and an open-source MCP server. Connect Claude Code, Cursor, or any agent you've built, and it reads and writes your whole business: projects, tasks, time, invoices, and contacts.
Every integration is an agent tool.
Connect a tool once and it shows up as a tool group in every agent config. Toggle read or read+write per agent, exactly the same way you scope native app tools.
Stripe
Slack
GitHub
WordPress
HubSpot
Salesforce
Notion
QuickBooks
Mailchimp
Klaviyo
Jira
LinearQuestions about agents.
How is an AI agent different from a regular chatbot?
A chatbot answers. An agent acts. Helm agents have scoped access to real tools across the platform: creating tasks, sending invoices, drafting emails, posting to Slack. Autonomy levels let you pick how much they decide for themselves. Memory lets them get better at your business over time.
How do I know an agent won't do something I don't want?
Three layers. First, tool access is scoped: agents only see tool groups you enable. Second, autonomy is configurable: start in Observe or Suggest mode and review every action before approving. Third, per-channel autonomy overrides let you be stricter on client-facing channels like email. Every action is logged.
Can agents delegate work to each other?
Yes. An agent can assign a task to another agent by name and include handoff instructions. The receiving agent picks it up and starts working automatically. That's how you chain specialists: a research agent gathers, then hands off to a copywriter agent to draft. Agents also sit on your org chart with roles, departments, and reporting lines, so structure and delegation line up.
Can multiple agents of the same role exist?
Yes. Create multiple Project Manager agents for different client tiers, each with different instructions, tool scopes, and spending limits. Or one Account Manager per major client. No unique constraint on role.
Where are agents reachable?
In-app chat, Slack channels and DMs, Email (Gmail connected), and scheduled runs. Each channel has its own permissioning. Invite an agent to a Slack channel with /helm invite. Assign it to a task and it runs. DM the workspace DM handler for cross-project work.
How much do agents cost?
Agents are included in every Helm plan. You set a monthly USD cost cap per agent; runs pause when the budget is hit. Underlying model cost is billed to your workspace, not per-seat. Heavy usage can be tuned with lower-cost model settings per agent.
Hire your first AI agent.
Every agent included. Per-workspace pricing. No per-seat tax.