CodeWords
Also known as: CodeWords AI, codewords.ai, Agemo AI, Agemo CodeWords
CodeWords, from Agemo AI, turns a plain language description of a recurring job into a deployed automation or AI agent: its builder agent Cody plans the work, writes the code, connects tools from more than 3,000 integrations, tests and debugs it, and runs it on a schedule, a trigger or a request. Workflows can drive a cloud browser, answer on WhatsApp and Slack, and be built from Claude, ChatGPT or any MCP client.
CodeWords is made by Agemo AI, Ltd. and is sold to people who can describe a job they repeat every week but will not maintain a flowchart or a code repository to automate it. Jumia, the ecommerce company, runs its JForce field agent operation on CodeWords, with a daily automation posting the best deals into JForce WhatsApp groups and seller support handled alongside it. The same builder also makes web apps and websites with a managed backend.
Cody, the builder agent, moves through chat, plan, build and debug modes: it researches, asks clarifying questions, presents a plan as a todo list for approval, then writes the service code, connects integrations through OAuth or stored API keys, registers triggers, tests each stage and fixes errors before deploying the workflow to its own run page. Every automation has a graph view and a code view, version history with restore, and stays editable by describing the change.
Workflows start on schedules, on Slack, WhatsApp, Gmail and Google Calendar events, on webhooks or through the API, and can call models from OpenAI, Anthropic, Google Gemini and Perplexity, scrape with Firecrawl, keep state between runs in Redis, and drive a cloud Web Agent browser or the user's logged in Chrome. Cody keeps a structured memory of each user's role, workflows, connected tools and recurring patterns, and the Claude connector, ChatGPT plugin and an MCP server let those assistants build and run CodeWords work directly.
The Free plan includes starter credits and limited WhatsApp use. Pro costs $39 a month or $390 a year with 60 monthly credits, API access, Claude and ChatGPT connectors and top ups at a 1.5x bonus, and Business costs $100 a month with 200 monthly credits, business templates, white labeled Cody and a 2x top up bonus; custom plans go through sales. Credits are spent on building, editing and running, and wallet top ups start at $10. Each run executes in an isolated sandbox on AWS that is destroyed afterwards, data is encrypted at rest and in transit, and SOC 2 Type II certification is in progress; there is no self hosted option.
CodeWords fits operators, founders, agencies and small teams who want recurring work automated through conversation, including WhatsApp and browser based tasks, without owning the code day to day. It fits less well for teams that need fixed costs, self hosting, single sign on and certified security today, or a flowchart they control node by node.
Vendor details
Canonical URL
https://www.codewords.ai
Category
Agent builder
Subcategory
Conversational builder where the Cody agent writes, deploys and maintains code based automations and AI agents across 3,000+ integrations, WhatsApp and a cloud browser
Funding status
Privately held; made by Agemo AI, Ltd.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
More than 3,000 integrations, among them Gmail, Outlook, Google Sheets, Drive and Calendar, Notion, Slack, HubSpot, Airtable, Shopify, GitHub, Instagram, LinkedIn, Stripe and WhatsApp, connected by OAuth with multiple accounts per app, plus any external API through stored keys. Built in services include Firecrawl scraping, Redis storage, a cloud Web Agent browser and a Chrome extension for logged in sites. Models come from OpenAI, Anthropic, Google Gemini (with Imagen and Veo) and Perplexity. Workflows are callable by REST API and webhooks, and a Claude connector, ChatGPT plugin and MCP server for Claude Code, Cursor and VS Code build and run work from those tools.
In practice
A field sales operation describes a morning job to Cody, which builds an automation that pulls the day's best deals each morning and posts them into the team's WhatsApp groups, while a second workflow answers seller questions in those groups.
A founder asks Cody for a personal assistant on WhatsApp; Cody plans the build, connects Google Calendar and Gmail, registers the WhatsApp webhook, tests it, and the assistant then reads and creates calendar events and drafts replies from her own number.
A recruiter has Cody build a weekly workflow in which the Web Agent signs in to a job board through a live cloud browser, collects new candidate profiles, enriches them with Perplexity research and writes the shortlist to Google Sheets, with failed runs visible in Activity.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
10.5 / 14 capabilities · 75%
| Integrations & Tool Calling | Full |
|---|---|
|
More than 3,000 integrations connect through OAuth from the Integrations settings, among them Gmail, Outlook, Google Sheets, Drive and Calendar, Notion, Slack, HubSpot, Airtable, Shopify, GitHub and Instagram, with separate Work and Personal accounts for the same app. Any other service joins through API keys stored securely and used inside workflows, and built in services cover Firecrawl scraping, Redis storage and LinkedIn messaging through the user's session. Because Cody writes real code for each workflow, an automation can read a lead from a form, enrich it, create the HubSpot record and message the owner on Slack in one run. Sourcedocs.codewords.ai/help-center/billing/integrations.mdread 2026-10-18 |
|
| Workflow Orchestration | Full |
|
Cody switches on its own between chat, plan, build and debug modes: complex requests with several integrations get a plan laid out as a todo list, then Cody writes the service code, configures tool connections, registers triggers, tests each stage and deploys. The result is a multistep workflow shown as a graph of nodes, from trigger through each step and integration to the output, with the code view behind it. AI agent services built this way combine several tools, a persona and their own triggers, and when Cody stays blocked after repeated attempts it pauses, explains and asks how to proceed. Sourcedocs.codewords.ai/automations/how-cody-works.mdread 2026-10-18 |
|
| Knowledge Grounding & RAG | Partial |
|
Workflows ground their output in live sources fetched at run time: Firecrawl scrapes websites into clean text and watches pages for changes, Perplexity supplies search grounded answers, and integrations read documents from Google Drive, Notion and Sheets. Cody itself researches with web search, documentation and page scraping before building. There is no knowledge base that indexes a company's documents for retrieval, and citations to stored sources are not part of the built agents unless Cody codes them. Sourcedocs.codewords.ai/automations/firecrawl-for-web-scraping.mdread 2026-10-18 |
|
| Human Oversight & Guardrails | Partial |
|
Oversight sits mostly at build time. Cody confirms requirements, shows its plan as a todo list that the user reviews, edits and approves before building, and asks for approval before a web app goes to production. LinkedIn sends and posts require the account owner's approval and configured rate limits, WhatsApp conversations can be handed to a person with full context, and unverified booking requests are captured for later human confirmation. Once a workflow is live, though, the builder offers no general approval step that pauses a run before a sensitive action, so any such gate has to be coded into the workflow by asking Cody for it. Sourcedocs.codewords.ai/automations/the-building-process.mdread 2026-10-18 |
|
| Security, Identity & Governance | Not documented |
|
CodeWords is in the process of obtaining a SOC 2 Type II report and holds no completed attestation or certification yet. Data is encrypted at rest and in transit on AWS, each run executes in an isolated sandbox that is destroyed when it finishes, Cody collects credentials through a secure flow instead of chat text, and internal access reviews, data classification and retention policies are in place, with a DPA available. On the customer side, workspaces organize work into projects and folders and workflows can be shared by link, but there are no roles, team permissions or single sign on for deciding who in a company can build, edit or run an automation. Sourcedocs.codewords.ai/security.mdread 2026-10-18 |
|
| Observability & Auditability | Full |
|
Activity brings past chats, workflow runs, scheduled runs and triggers into one view, with runs searchable and sortable by date or cost and filterable by status and by whether a schedule, a person or a trigger started them. Each run page shows the step in progress as it updates, the output and the error details of a failed run. Usage breaks credit spend down by day, by chat and by workflow run, and the graph view and code view expose exactly what each automation does. Sourcedocs.codewords.ai/automations/activity.mdread 2026-10-18 |
|
| Memory & State Persistence | Full |
|
Cody builds a structured memory of each user from six sources: account basics, onboarding context, a live view of the user's workflows and how they perform, the names of connected services, behavioral patterns distilled from periodic reviews of conversations, and recent chat topics, each refreshed on its own schedule, with credentials never stored. That summary lets Cody skip setup questions in new chats, and it can also search past conversations to recall earlier work. Inside deployed workflows, Redis stores and recalls data between runs, so a bot can remember what it already posted or who it already answered. Sourcedocs.codewords.ai/automations/memory-in-codewords.mdread 2026-10-18 |
|
| Deployment & Data Residency | Not documented |
|
Workflows run only in CodeWords' managed cloud on AWS, each in a sandbox created for the run and destroyed afterward. CodeWords says plainly that engineering teams wanting to self host the runtime are not the audience it builds for, and buyers get no choice of hosting region or data residency option. Web apps built on CodeWords can be exported to the customer's own GitHub repository, and the code behind each automation belongs to the customer, but automations themselves stay on the managed platform and cannot be moved into a company's own cloud. Sourcecodewords.ai/llms.txtread 2026-10-18 |
|
| Prebuilt Agents / Templates / Packs | Full |
|
More than 876 ready templates can be remixed instead of built from scratch, grouped by job across analytics, sales, marketing, support, operations, finance and more, from a competitor news monitor and customer feedback analyzer to trading signal bots and daily market briefings. WhatsApp alone has more than 125 templates for personal assistants, group bots and lead qualification, Business plans add a set of business templates, and Cody adapts any template to a team's own tools by conversation. Sourcecodewords.ai/templates/llms.txtread 2026-10-18 |
|
| Triggers & Channel Coverage | Full |
|
Schedules run a workflow once or on a recurring timetable in the user's timezone, with optional email confirmations, and triggers start one on Slack messages, reactions or channel joins, WhatsApp DMs, group messages and reactions, new Gmail emails, label changes or drafts, and Google Calendar events. Webhooks give each workflow or agent an HTTP endpoint for outside systems, the API runs workflows synchronously or asynchronously with file uploads, and runs can also start from chat or a run page. On WhatsApp, a workflow can answer on the customer's own number through the personal device route without a Meta Business API application. Sourcedocs.codewords.ai/automations/schedules-and-triggers.mdread 2026-10-18 |
|
| Model Flexibility & Routing | Full |
|
Workflows can use Anthropic's Claude Opus, Sonnet and Haiku, OpenAI's GPT 5, GPT 4.1, o3 and o4 mini, Google Gemini with Imagen 4 for images and Veo for video, and Perplexity for search grounded answers, with Cody choosing the model and wiring the calls. Users can ask Cody to switch a step to a cheaper model to cut credit use, and any other model provider can be reached through its API key. Templates name the model they run on, such as GPT 4.1 or Claude Sonnet 4.5. Sourcedocs.codewords.ai/llms-full.txtread 2026-10-18 |
|
| APIs / SDKs / MCP Extensibility | Full |
|
Paid plans get a CodeWords API key for calling workflows from outside, synchronously or asynchronously with file uploads, alongside webhook endpoints and an OpenAPI description on the developer pages. An MCP server lets Claude Code, Cursor, VS Code and other MCP clients build agents and automations on CodeWords, and the Claude connector and ChatGPT plugin let those assistants write the code, connect Slack, Gmail or Notion, deploy, schedule runs, read logs and fix a failed run without leaving the chat. Everything Cody builds is real code the customer owns. Sourcedocs.codewords.ai/automations/calling-codewords-workflows.mdread 2026-10-18 |
|
| Testing, Debugging & Optimization | Partial |
|
Cody tests integrations and APIs before building, tests each stage as it builds and fixes errors on its own, and a debug mode takes over when a user reports a broken workflow. Triggers can be tested with real or sample data, the graph view shows which stages passed, and version history previews and restores earlier versions of an automation. There are no saved test sets, scored evaluations of an agent's answers or comparisons between versions. Sourcedocs.codewords.ai/automations/how-cody-works.mdread 2026-10-18 |
|
| Browser / Computer-use | Full |
|
The Web Agent is a cloud browser that works like a person, navigating pages, clicking buttons, filling forms, scrolling to load content and working behind logins, while the user watches through a live browser view or comes back to the results. The Chrome extension runs tasks inside the user's own logged in browser for private pages such as LinkedIn feeds and dashboards, including clicks and form fills on scheduled runs, with a five minute limit per run. Cody chooses between the Web Agent and Firecrawl depending on whether a page needs interaction. Sourcedocs.codewords.ai/features/web-agent.mdread 2026-10-18 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Pro $39 a month, or $390 a year ($32.50 a month)
Monthly or yearly plan with a monthly credit allowance plus prepaid wallet top ups; credits pay for building, editing and running workflows
Included quota
Free: starter credits, 100 workflow runs a month, WhatsApp 5 DMs a day and 1 linked group, no API access. Pro: 60 credits monthly, 2,500 workflow runs a month, credit roll overs, custom domains, WhatsApp 20 DMs a day and 5 groups. Business: 200 credits monthly, unlimited workflow runs, credit roll overs, business templates, white labeled Cody, WhatsApp 50 DMs a day and 10 groups.
What is public
Plan prices, monthly and yearly rates, monthly credit allowances, WhatsApp limits, top up presets and bonus multipliers.
Billing mechanics
Self serve subscription billed monthly or yearly through Stripe, with a separate prepaid wallet, optional auto top up and per day usage reporting.
Cost watchouts
Building and editing with Cody consume credits too, and browser automation and expensive models cost more per run. API access and the Claude and ChatGPT connectors need a paid plan.
Variable cost rationale
Credits are spent on building and editing as well as running, and use varies with workflow complexity, model choice, browser automation and run frequency; the company itself warns that costs are not fixed in advance, so heavy or frequent workflows lean on wallet top ups.
Additional watchouts
Long builds and frequent edits spend credits before a workflow ever runs in production.
Overage / add-ons
Pro and Business buy wallet top ups in $25, $50, $100 or $300 presets or any amount from $10, with a 1.5x credit bonus on Pro and 2x on Business, and can turn on auto top up; top up credits stay until used. Free plans cannot top up.
Sales call required
No, self serve available
Free / trial
Free plan with starter credits; the first workflow is free to build
Lowest paid plan
Pro at $39 a month or $390 a year: 60 monthly credits, API access, Claude and ChatGPT connectors, 1.5x top up bonus
Commercial notes
Business adds business templates and white labeled Cody for agencies; custom and individually scoped arrangements go through sales.
Key ambiguities
The value of one credit and how many runs 60 or 200 monthly credits cover are not stated, so monthly cost depends on top ups.
Missing data
Credit unit value, credits included on Free, and custom plan pricing.
Related vendors
- Lindy — Lindy is an AI teammate that lives in Slack, answers from your tools…
- Activepieces — Open source, MIT licensed automation platform where every one of 700…
- Agenta — Agenta is an open source workspace where teams build AI agents by…
- AgentX — No code platform to build, evaluate, and deploy multi agent…
- AI Library — No code platform for building and running AI agents, with a…
- Aigensei — Enterprise agentic workflow platform that runs business processes…