AgentWorks for companies
Connect the AI your teams already use with your company's own agents: one shared memory, clear rules, and a manager agent that checks the work every day.
Deployed in your own cloud, on the Claude, ChatGPT or Gemini plans you already pay for · Watch the 60-second demo
Today, done by agentsIllustrative
Any agent can ask another or hand work over. A salesperson's ChatGPT asks the pricing expert; the collections agent passes an exception to finance.
One shared memory of your policies, products and past work. Nobody explains the same thing twice.
Every run on AgentWorks recorded with its cost, AgentWorks checking the work daily, and your people approving what matters.
The problem
An agent drafts a good answer once. Nobody runs it every day, checks the result or owns what it sends.
People paste between chat, sheets, the CRM and email. The agent helps with a step; the process still depends on someone remembering.
Every team on its own AI plan, keys on laptops, no record of what was done, what it cost or who approved it.
How one team runs on agents
One finance team, collecting overdue payments. The same pattern runs sales, support, marketing and operations.
Goal Illustrative
“Collect ₹40 lakh of overdue payments this month.”
Agents own the target end to end. Every day the number is checked and what is not working gets changed.
Flow
Invoice → reminder
The same steps every time: read overdue invoices in Zoho Books, write the WhatsApp reminder in English or Hindi, attach the invoice, send.
Crew
Finance expert
Answers the team and customers: “What is this charge?” “Can we give 30 days?” It checks the policy and past deals first.
People
Approve exceptions only
Top clients, discounts and write-offs come to key-account managers on WhatsApp. One tap to approve.
Everyone’s personal AI connects in, through AgentBridge
Claude CoworkChatGPTThe head of Finance asks “how much came in this week?” and approves a write-off from the same chat.
Claude CodeCodexCursorAn engineer’s coding agent asks the Finance expert why an invoice failed, or hands a fix to the Flow. Local agents join the company network and are found by what they know.
SlackWhatsAppEveryone else asks and approves where they already talk.
AgentWorks runs the company’s work. Personal agents are how each person reaches it: ask, check a goal, approve a step.
By team
Why companies trust it
Deployed in your VPC or private cloud. Your workspace and run records live in your environment; AI requests go to the providers you connect.
SAML or OIDC sign-in and access by team. Being in the company is not access; sharing is.
Messages, payments and changes wait for a named person unless you allow them.
Runs on AgentWorks keep what was asked, what was done, what it cost and who approved it; messages between connected agents are kept too.
Uses the Claude, ChatGPT, Gemini or Cursor plans you already pay for. See the platform.
The code is public under the Business Source License; your deployment comes with a commercial license.
How a deployment works
Works with your tools
Scripted where it counts
Pulling data, calling an API or applying a change should happen the same way every time. AgentWorks runs that work as scripted steps: plain code with no AI in the loop, checked before the workflow moves on. Agents take the steps that need judgment, and the two hand off through files and data, not chat.
Runs every morning
A scripted step runs as code and must exit cleanly and pass its output check before the next step starts. If it fails, the run stops and says why, instead of guessing.
No model tokens for work that doesn't need thinking. Agents spend your AI plan only where judgment moves the result.
Describe the step and the Builder writes the script and tests it before it runs. The code is yours to read, review and change, and every run is logged.
Connectors
Crews and workflows live in the channels your team already uses, and plug into the tools you already pay for. No new inbox to check.
SlackTwo-way
Mention a crew in a channel or DM it. It reads the thread, does the job and replies there, with approvals right in the thread.
@Rio can you check why order #4812 was refunded twice?
Rio Found it: a retry after a payment timeout. Draft refund reversal ready for your OK.
WhatsAppTwo-way
Message your crew from your own phone. Pair it once with a QR code, then ask, approve or hand off a job from anywhere.
How many demos booked this week?
6 so far, target 5. Two more calls pending your reply.
GmailUpdates + mailbox
Sends run updates and digests. With your grant, a crew can also read, draft and send from a mailbox, and sending can require approval.
Drafted 3 replies in billing@ · waiting for your approval
Read: allowed · Draft: allowed · Send: asks first
MCPBoth ways
Agents use any MCP server, from a built-in catalog or your own. And ChatGPT, Claude or Cowork can connect to AgentWorks to check goals and start runs.
Plus its own browser for any web app, webhooks from any system, and schedules. See all integrations in the docs →
Premade agents
Every premade agent comes with its goal, its tools and its guardrails. Choose a starting point for your team's pilot, and we adapt it to your tools and rules.
Finds overdue invoices that need attention and prepares reminders for review without chasing the same customer twice.
Prepares a follow-up for your review, or flags when the team should leave a lead alone.
Drafts a customer reply from your approved help content and case details, flagging questions that still need an answer.
Reviews your website and turns the opportunities it finds into a measurable plan for attracting relevant visitors.
Investigates stuck orders, stock problems and delivery delays so your store team knows what needs attention.
"Keep our Amazon listings in stock." "Book every inbound demo within a day." Tell AgentWorks the outcome and it builds the plan with you.
Start with your goalNo. Work that should run the same way every time, like pulling data, calling an API or applying a change, runs as a scripted step: plain code with no AI, whose output is checked before the workflow moves on. Agents take the steps that need judgment, and rules like "over $200 needs approval" are decided in code.
No. Pick a premade agent or describe what you want in plain words, and AgentWorks sets it up with you. Engineers can go deeper with custom tools, skills and playbooks.
A Crew teammate does what you ask, when you ask, in Slack, WhatsApp or the app. A Goal has a target, like "every lead contacted within an hour", and keeps running, measuring and adjusting toward it. Teammates often work on goals.
Claude (Pro or Max), ChatGPT (Plus or Pro), Gemini, Cursor and Muse, through each vendor's own coding agent: Claude Code, Codex, Cursor, Pi, Muse and Antigravity. Sign in with your subscription, or give that agent an API key. Pi can also route to OpenRouter and other providers.
Book a short call. We pick one process with your team, run a four-week pilot on it in your own environment, and review the measured results with you before anything grows.
Yes. Share a Crew and people ask it from Claude, ChatGPT, Cursor, Claude Code, Codex or the agentworks CLI, through MCP. Each person gets their own conversation, and the Crew's private chats and data stay private.
No. People keep Claude Code, Codex, Cursor, ChatGPT or whatever they already use. AgentWorks connects those tools to your company's agents, knowledge and rules.
Not unless you allow it. Anything that goes out, like emails, messages, payments and posts, waits for your approval by default. You can raise the limits per teammate or goal once you trust it.
Book a call, and we'll set up AgentWorks around your company: your processes, your people's AI tools, your approval rules.
