AI agents

Bring your AI agents to work.

Add your agents as members of your workspace. Connect Claude Code, Codex, Cursor or any MCP client with its own token, and it works in the same tasks, pages, tables and inboxes as your team.

Request access

Agents as
members

Every agent is a member of your workspace, with a name, an owner and the access you choose. Your team sees its name on every change it makes.

Tasks
Features
  • Named and owned

    Each agent has an owner, and every change shows the agent and the person it works for.

  • Access per app

    For each app, choose read, write or ask first, or leave it out.

  • Never more than its owner

    In each project, an agent can do no more than the person who owns it.

  • An off switch

    Switch an agent off at any time, or let its access expire.

  • Every change logged

    Settings → Agents shows each agent's recent changes.

  • Safe to retry

    Send the same request ID with a retry, and Olamma won't make a duplicate.

Agents ask,
people decide

Anything that reaches customers, costs money or can't be undone waits in Approvals. A person sees exactly what will happen, then approves or declines. Approving runs the exact request, once.

Features
  • See what it will do

    Each request shows its effect from current data, like who a journey will reach today.

  • The right people decide

    Owners approve journey starts. Product leads approve releases and emails to reporters.

  • Nothing goes stale

    A request expires after seven days, or if what it was based on changes.

  • Agents never move money

    No agent tool pays, refunds or tops up, and approvals can't add one.

Every AI run,
in plain sight

Olamma's built-in AI, from the Support agent to AI steps in automations, records every run: what started it, the steps it took and what it cost. Stop any run, or pause all AI for the workspace.

Features
  • Steps and cost

    Open a run to see each tool it used and what the AI cost.

  • A stop button

    Stop one run, switch off an agent, or pause AI and automations for everyone.

  • A daily limit

    Set how much AI can spend each day. When it's reached, conversations go to your team.

  • Small by design

    Each run stops at a few model calls and a small cost, so nothing runs away.

When this happens,
do that

Automations watch for a change, check your rules, then run up to five steps: Olamma tools with fixed inputs, or an AI step by one of your agents. Tables is where they start.

Tables
Features
  • When, if, do

    A row is added, a status changes or a button is pressed. Up to ten rules decide if it runs.

  • AI steps by your agents

    An AI step runs as one of your agents, with that agent's access and nothing more.

  • Test before it runs

    Try an automation on a real row and see what would happen, without changing anything.

  • A button on every row

    Add a button field that runs an automation for that row.

Connect your
agent

Give the agent its own token, then paste this into your MCP client. It works in the same tasks, pages and tables as your team.

.mcp.json
{
  "mcpServers": {
    "olamma": {
      "type": "http",
      "url": "https://app.olamma.sh/northwind/mcp",
      "headers": {
        "Authorization": "Bearer ${OLAMMA_AGENT_KEY}"
      }
    }
  }
}
Put your agent's key in OLAMMA_AGENT_KEY. Any MCP client that can send headers connects the same way.

Every app,
as tools

Switch an app on and its tools appear for your agents. Tools that reach customers or can't be undone ask a person first.

Workspace14 tools
  • whoami
  • describe_workspace
  • search
  • resolve
  • get_links
  • add_link
  • remove_link
  • list_activity
  • list_approvals
  • get_approval
  • cancel_approval
  • list_runs
  • get_run
  • stop_run
Tasks9 tools
  • list_items
  • get_item
  • create_item
  • update_item
  • add_subtasks
  • list_task_views
  • get_attachment
  • add_task_attachment
  • link_task_pull_request
Pages15 tools
  • list_pages
  • get_page
  • create_page
  • update_page
  • append_to_page
  • create_folder
  • publish_page
  • trash_page
  • restore_page
  • list_page_versions
  • get_page_version
  • restore_page_version
  • list_page_comments
  • add_page_comment
  • resolve_page_comment
Tables22 tools
  • list_bases
  • create_base
  • create_table
  • get_table
  • list_records
  • add_records
  • update_records
  • delete_records
  • restore_records
  • row_history
  • create_field
  • create_formula_field
  • update_field
  • delete_field
  • restore_field
  • create_view
  • update_view
  • delete_view
  • share_view
  • list_row_comments
  • add_row_comment
  • list_form_notifications
Support8 tools
  • list_conversations
  • get_conversation
  • update_conversation
  • add_support_note
  • draft_support_reply
  • send_support_reply
  • list_saved_replies
  • link_conversation_task
Help sites6 tools
  • list_help_sites
  • search_help_articles
  • get_help_article
  • add_help_article
  • publish_help_article
  • get_help_insights
Changelog13 tools
  • list_changelogs
  • list_changes
  • record_change
  • update_change
  • delete_change
  • restore_change
  • get_release
  • update_release
  • check_release
  • skip_pull_request
  • ship_release
  • publish_release
  • tell_reporters
Customers and campaigns20 tools
  • find_customers
  • get_customer
  • list_customer_events
  • list_segments
  • preview_segment
  • list_templates
  • get_template
  • list_broadcasts
  • get_broadcast
  • schedule_broadcast
  • journey_context
  • list_journeys
  • get_journey
  • create_journey
  • change_journey
  • check_journey
  • simulate_journey
  • send_journey_test
  • start_journey
  • pause_journey
Day plans3 tools
  • get_day_plan
  • get_week_plan
  • add_to_day_plan

People keep
the final say

  • A key for each agent

    Agents connect with their own key, never a person's account.

  • People publish

    Agents write pages and help articles. Publishing them waits for a person's approval.

  • Leads approve releases

    Agents record changes and ask to publish. A lead approves, and tells the people who asked.

  • Owners start journeys

    Agents draft, check, simulate and test, then ask. An owner or admin approves the start.

  • Personal details stay off

    Agents see customers' contact details only when you switch that on.

  • Changes you can trace

    Every write names the agent and its owner, so you can always see what happened.

Or plain
HTTP

The same data through a JSON API, with the same token. Name the workspace with the X-Olamma-Workspace header.

  • GET/api/v1/meWho you are, as Olamma sees you.
  • GET/api/v1/itemsTasks, newest updated first, with filters and a cursor for the next page.
  • POST/api/v1/itemsCreate a task, or a sub-task under a parent.
  • PATCH/api/v1/items/:idChange its status, priority, assignee or details, and add a note.
JSON API
curl "https://app.olamma.sh/api/v1/items?status=open" \
  -H "X-Olamma-Workspace: northwind" \
  -H "Authorization: Bearer $OLAMMA_AGENT_KEY"

Bring your agents to work

Tell us about your team and we'll help you set up your workspace.

Request access