Agent Platform

AI agents that work where your team works.

Agents in Slack, WhatsApp, Telegram, email and on the phone that connect to your tools, follow your rules, and get real work done.

# finance-opsSlack
You@Manifest every Monday, find invoices more than 30 days overdue and send each customer a friendly reminder.
Manifest agent Here is the plan. Nothing runs until you approve it.
TRIGGEREvery Monday at 9:00
FINDInvoices 30+ days overdue, from your billing system
DRAFTOne reminder email per customer
APPROVEYou approve the batch here before anything is sent

How it works

From a sentence to a running automation.

01

Connect

Install the Slack app, connect a WhatsApp Business number, a Telegram bot, email, a phone number or the web widget. Then connect GitHub, Linear, Zoom or any MCP server.

02

Describe

Ask the team agent for anything. For repeatable work, the Automation Builder plans it with you, builds it, and tests it with real messages.

03

Run

Start on a message, a schedule, a webhook, an email, a call or an event. Sensitive actions wait for a person to approve the exact action.

04

Watch and improve

Every input, decision and tool call is on the activity timeline. Re-run past conversations against a change before you promote it.

Automation Builder

Describe it. Approve the plan. It runs.

Say "automate this". The builder asks what it needs to know, shows you a plan, waits for your yes, then builds the workflow, its triggers and its connections, and tests it before it goes live.

For one-off requests, just ask the team agent. It is always on, answers from your team's knowledge, and hands bigger jobs to the right workflow in a new thread.

UNDERSTANDAsks the questions that matter: which system, which customers, what counts as overdue.
PLANWrites the plan in plain language, with the tools and permissions it will need.
APPROVEWaits for you. Change anything, or approve it as is.
BUILDCreates the workflow, its triggers and its integrations.
TESTSends real test messages through it and checks the results.
RUNGoes live, with every run on the record.

Channels and triggers

Meet people where they already talk.

SlackDMs, mentions, threads and approval buttons
WhatsAppYour own WhatsApp Business number
TelegramDMs, groups, topics and commands
DiscordServers and threads
EmailA dedicated address for every workflow
PhoneInbound and outbound calls, transcripts, hand-off to a person
Web chatOne line of code on your site
REST APIStreaming replies and structured output
WebhooksAny system can start a run
SchedulesOnce, every few minutes, daily or cron

Events from GitHub, Linear and Zoom can start work too: a pull request, a delegated issue, a finished meeting recording.

Tools and integrations

Agents that can actually do the work.

Connect anything with MCP

Point the agent at any MCP server by URL, with API key, OAuth or signed-token auth, and choose exactly which tools each workflow may use. No server yet? Describe the API and the agent writes one; we host and log it for you.

GitHubLinearZoomGrafanaAWSFly.ioAny MCP server

Built-in capabilities

Real browser
Logs in, fills forms, uploads files
Private workspace
Files and code, per conversation
Web research
Search and read pages
Transcription
Speech to text
Phone calls
The agent places calls
Code sessions
Claude Code on your repos

Answers from what your team knows

Add your website, documents and data feeds. Agents search them, cite their sources, and can compile them into living wikis. Access follows role and channel.

Website crawlUploadsAPI sourcesAgent notesWikis

Trust and control

Autonomy you can see, and stop.

Approvals bound to the action

A yes in Slack approves the exact values shown, once, and expires. Change the action and it asks again.

Roles and permissions

Admins configure, members use, customers get served. Override access per workflow, per channel and per tool.

Encrypted secrets

Keys and tokens are encrypted per team, never shown back, and every use is audited.

Everything is an event

Every message, decision, tool call and reply lands on the activity timeline, with usage metrics you can export.

Models

Your choice of model. Your keys, if you want.

Pick a model per workflow from OpenAI, Anthropic, or hundreds more through OpenRouter, and set how hard it should think. Bring your own keys and they stay encrypted on your team.

OpenAIAnthropicOpenRouterBring your own key

Change with confidence

Test a change on real history first.

Every workflow is versioned. Eval Mode replays past conversations against new instructions or a new model, answers tool calls from the recorded run, and scores the results before anything reaches production.

For developers

An API, an MCP server, and a Claude Code plugin.

Run any workflow over REST with streamed replies and JSON-Schema output. Manage your agents, runs and knowledge from Claude Code through our MCP connector.

# Run a workflow and stream the reply curl -N "https://agents.manifestautomation.com/a/acme/agents/api/post/?stream=true" \ -H "Authorization: Api-Key $MANIFEST_API_KEY" \ -H "Content-Type: application/json" \ -d '{"request_data": {"message": "Summarize the open support tickets"}, "response_schema": {"type": "object", "properties": {"summary": {"type": "string"}}}}' # Or manage it all from Claude Code claude mcp add --transport http manifest-agents \ https://agents.manifestautomation.com/mcp

What would you hand to an agent first?

We onboard every team directly. Tell us the process and we will show you the plan.

Talk to us