GrowholdGrowhold guide
Growhold is a workspace for researching a market, planning products and features, and producing the documents that go with that work. This guide explains each part of the product, what it is for, and how to use it.
Getting started
How the parts of Growhold fit together. Read this section first if you are new. The sections after it cover each part in more detail.
Create a workspace
Sign up and Growhold creates a workspace for your company. The workspace holds your conversations, agent runs, documents and market data in one place, so everything you learn about a product is available to everything you do next. Every workspace has access to the full catalogue of managed agents from the start. During onboarding Growhold reads your company website and creates a first set of specialists from it, so early answers already reflect your products, tone and positioning. You can edit or replace these at any time.
Start in Chat
Chat is for research and planning work that does not yet have a fixed shape: understanding a market, comparing competitors, working out what a feature should do, or testing a positioning idea. It answers from your workspace knowledge, the specialists in scope and the market demand database, so answers are about your situation rather than general advice. When a question turns into a defined piece of work, Chat can start an agent for you.
Check demand in Market
Market is a database of live buying activity: what buyers in each industry are currently paying to have solved, which categories are growing, and where few suppliers exist. It lets you check a product direction against real demand before investing in it. Any demand pattern can be opened in Chat, where you combine it with your own knowledge and decide whether it applies to you.
Produce a deliverable with an agent
An agent produces a specific deliverable through a fixed sequence of steps: a screen plan and the mockups of the screens that matter, a competitive analysis, a positioning document, a feature brief. Use an agent when you know what you need and want it produced the same way every time, with each step building on the previous one. Visual steps deliver images, so a product can take shape on screen before anything is built. Every run appears in Work, where you follow progress and review the result.
Agents are grouped by the stage of building a product they serve:
- DiscoverDiscover ideas worth building
- SolveSolve the right problem
- ShapeShape it into a product
- DeliverDeliver work that ships
Each group on the Solutions page lists its agents with the steps they run and a sample outcome from a real run, so you can see the deliverable before starting one. Solutions →
Group related work into a build
A build represents one thing you are working on: a product, a feature or a concept. Assign chats and agent runs to a build and they share knowledge and memory with each other but not with other builds. Research about two different products stays separate, so search, memory and results remain accurate for each.
Knowledge accumulates
Files you upload, results you keep and facts learned in conversation become workspace knowledge. Later chats and runs use it, so the workspace becomes more accurate about your company over time. You decide what is kept. Results from a direction you abandoned can be left out so they do not influence later work.
Chat
Chat is where research and planning happen before the work has a defined output. It uses your workspace context, and it hands off to an agent when the next step is clear.
What Chat is for
Use Chat to research a market, compare competitors, plan a product or feature, draft requirements, or test a business decision before committing to it. Answers draw on workspace knowledge, the specialists you have attached, the build you are in, and the live market database. This is the difference from a general assistant: Chat already knows your company, your products and your market. Agents produce a known deliverable. Chat is for working out what that deliverable should be.
Scope: what a conversation can see
The Scope column on the right controls what a conversation draws from. Selecting a build files the conversation under it, adds the build’s knowledge to answers, and stores what the conversation learns as memory for that build. Source toggles add or remove market intelligence, your knowledge base and past work results. Attaching a specialist adds its instructions and documents. You can change the build during a conversation. A marker in the transcript shows where the change happened, and everything after it belongs to the new build.
Memory
Growhold keeps the full transcript of each conversation and also extracts key facts from it: decisions made, constraints identified, positions taken. A new chat in the same build starts with those facts, so you do not repeat context. Memory is stored per build, so facts about one product are never applied to another.
Cards from Chat
When a conversation reaches a conclusion worth keeping, ask Chat to create a card. A card is a structured document with sections, tables, charts and generated images where you ask for them. It can be edited, versioned, exported and shared by link.
Handing work to an agent
For work that needs several steps, ask Chat to start an agent. It picks a suitable agent from the catalogue, starts the run with the conversation as its brief, and the run’s progress appears right in the thread. When it finishes, the outcome is one click away in Work, and Chat can keep working from it.
Work
Work lists every agent run in the workspace, running or finished. Use it to follow progress and to decide which results become part of your knowledge.
The work list
Each entry shows the agent, the build it belongs to, its status and how many steps have completed. Runs in progress update live. Filter by agent, status or tag, or scope to a build to see only the runs for one product.
Keeping results
A finished run is a set of cards, one per step. Keep a card and it is added to the knowledge base, where later chats and runs in the same build can use it. Leave it out and it stays readable in the run but does not affect other work. This keeps the knowledge base accurate: a competitive analysis from a direction you dropped should not shape the next product brief.
Builds
A build is one product, feature or concept. It groups the conversations, runs and knowledge about that idea so they inform each other without mixing with other work.
What a build is
Most teams work on more than one thing at a time. Without separation, a search for pricing research returns results from every product, and memory from one project influences another. A build gives each idea its own conversations, runs, documents and memory. The builds page lists each build with the deliverables it has produced and a fourteen-day activity chart, so you can see which builds are active. Create a build when an idea needs more than one conversation.
Working in a build
Opening a build shows its activity feed with recent chats and runs beside it. The Scope column on Chat and Work sets which build you are working in and keeps that selection as you move between pages. The selected build determines which knowledge is searched, which memory is recalled, and which conversations and runs are shown.
Cards
A card is a single structured document produced by an agent step or by Chat. An agent’s deliverable is the set of cards from all its steps.
What a card is
Each agent step produces one card: a competitor analysis, a mood board, a content calendar, a set of recommendations. Cards use defined sections, tables and key figures so they can be read quickly and compared across runs. When you keep a card it becomes part of the knowledge base for its build, and later chats and runs can draw on it.
Editing and version history
Cards can be edited after delivery, either by describing the change and letting AI apply it or by editing the content directly. Each edit saves a new version, so earlier drafts remain available and any change can be reverted. Regenerate a card when the underlying context has changed, for example after new research has been added to the build.
Images and charts
Card templates can include image and chart placeholders. An image placeholder generates a visual as part of the card: a mood board, a UI mockup, a concept illustration. A chart placeholder renders pie, bar, line, area or heatmap charts from data produced in the run, so quantitative findings are shown as charts rather than tables of numbers.
Editing a generated image
Generated images can be edited without regenerating the card. Open the editor on the image and describe the change: a different color palette, a plain background, an element removed. Mask a region to limit the change to that area and leave the rest untouched. Each image edit creates a new card version.
Exporting cards
Delivered cards are shown on the run page grouped by step. Export them for use outside Growhold as reports, briefs, plans or creative assets.
Market
Market is a database of current buying activity by industry. It exists so product decisions can be checked against what buyers are actually paying for.
What Market shows
Market shows demand patterns: recurring cases of a specific type of buyer, in a specific industry, paying to have a specific problem solved. Patterns are built from purchasing signals rather than surveys or opinion. Demand is grouped into three areas by what is being bought: products to build, workflows to automate, and services to deliver. Within each area, patterns are grouped by vertical so you can follow the industries relevant to you.
Reading a pattern
Each pattern has a strength rating based on how many signals match it, how consistently buyers describe the problem, and how active the category has been recently. Strong patterns are confirmed, active demand. Weak patterns are often early trends or narrow niches, which can be an opening if you can serve them. Buyer intent tags show the words buyers themselves use, which helps when naming features and writing copy that matches how customers search.
From a pattern to a deliverable
Open any pattern in Chat to analyze it with your own context: who the buyers are, whether your product or planned feature addresses the problem, and what a competitive offer would look like. From there, run an agent to produce a competitive analysis, positioning document or feature brief, scoped to the relevant build. Chat and agents query the full demand database, not only the patterns on screen.
Agents
An agent is a fixed sequence of steps that produces a specific deliverable. Growhold maintains a catalogue of managed agents, and you can copy any of them to customize.
Agents are grouped by the stage of building a product they serve:
- DiscoverDiscover ideas worth building
- SolveSolve the right problem
- ShapeShape it into a product
- DeliverDeliver work that ships
Each group on the Solutions page lists its agents with the steps they run and a sample outcome from a real run, so you can see the deliverable before starting one. Solutions →
What an agent is
An agent defines the steps needed to produce one deliverable, such as a competitive analysis, a feature brief or a content calendar. Each step produces one card, and steps run in order so later steps can use what earlier ones produced. Most agents in the catalogue are managed by Growhold: shared across workspaces, read-only, and updated centrally so their behavior stays consistent with this documentation. You can run a managed agent, scope it to a build and attach specialists, but not edit it. To change the process itself, copy the agent. The copy belongs to your workspace, can be edited freely, and no longer receives updates from the managed version.
Finding or creating an agent
Browse the catalogue by name or category, including competitor analysis, SEO audit, content calendar and product brief. If no agent matches, describe the deliverable you need and Growhold creates an agent with a name, instructions and steps. Copy an existing agent when the process is close to what you need, and create a new one when your team follows a specific process of its own.
Running an agent
Start a run from the agent’s page on Agents, or ask Chat to do it: describe the work and Chat selects the agent and starts the run. In both cases, choose the build the run belongs to and the specialists that should contribute. Then choose the mode. A conversational run goes through the steps with you, so you can review and redirect each card as it is produced. An autonomous run completes all steps without input and notifies you when finished. Use conversational mode when the direction still needs decisions, and autonomous mode when the brief is clear. Runs appear in Work.
How an agent is built
An agent is a list of steps. Each step has its own instructions, input fields that must be filled before it starts, and a card template in markdown that defines the output structure: sections, tables, statistics, and placeholders for images and charts. Because steps run in order, a specification step can depend on a requirements step having completed first.
Connections
Connections give agents access to live external data: web search, page scraping, news feeds, social media and video platforms. Enable each one per agent. With a connection active, the agent gathers current information during the run rather than relying only on workspace knowledge. This matters for competitive analysis, market monitoring and content research, where the source material changes often.
Specialists
A specialist holds knowledge about a domain: your brand, your product, your customers. Attach one to a conversation or run and its instructions and documents are applied.
What a specialist is
Agents define a process and are shared across workspaces, so they cannot contain your company’s specifics. Specialists are where those specifics live. A design specialist carries your palette, voice and image rules. A product specialist carries your roadmap and positioning. Attaching a specialist to a run makes a shared agent produce output that follows your standards.
Creating a specialist
Describe a domain and Growhold generates a specialist for it, or give it a URL and it builds one from the page content. Onboarding creates an initial set from your company website. After creation you can edit the name, instructions, tags and image.
What a specialist contains
A specialist has three parts. Its guide is a set of instructions on how to think and write: brand voice, naming conventions, document structure, review rules. Its knowledge is documents: strategy decks, product specs, customer research, competitive analysis. Its visuals are palettes, sample images and design references used when generating images. All three apply wherever the specialist is attached.
Knowledge
Knowledge is the set of documents and results your workspace can draw on. Each item belongs to one of three scopes, which control where it applies.
Three scopes: workspace, specialist, build
The Knowledge page has two tabs. Knowledge Base holds documents your team has added. Work Results holds the cards you chose to keep from agent runs, grouped by run. Every item in the Knowledge Base belongs to one scope. Workspace knowledge applies to everything: facts about your company that hold regardless of project. Specialist knowledge belongs to one specialist and applies wherever that specialist is attached. Build knowledge applies inside one build. The scopes combine: a chat in a build uses workspace knowledge, the knowledge of any attached specialists, and the build’s own knowledge. Search works by meaning rather than exact keywords, across both tabs.
Adding knowledge
Upload files or paste text: PDF, Word, markdown, plain text and other common formats. For material that arrives regularly, such as customer feedback or reports, forward it by email or send it to your workspace webhook and it is added automatically. Useful sources include product specs, brand guidelines, market research, competitor pages, customer feedback and meeting notes.
Assigning a scope
New items wait in a queue until you assign them a scope. Assign to the workspace when the content should apply to all work, to a specialist when it belongs to that domain, and to a build when it concerns one product. An item has exactly one scope. Moving it to a new scope removes it from the old one, so knowledge does not continue to influence work it no longer belongs to.
Plans and billing
Every plan includes all features and unlimited users. Plans differ only in the number of credits per month.
Plans
Three plans are available: Starter at $49 per month with 2,000 credits, Growth at $99 per month with 5,000 credits, and Scale at $199 per month with 12,000 credits. All plans include unlimited users, all features and the full agent catalogue. Adding a payment card gives 500 free credits to start. Choose a plan when they run out.
Usage and subscription
The plans page shows your credit balance and what consumed it, so you can see how credits were spent before changing plan. Upgrades, downgrades, payment methods and invoices are managed in the billing portal.
Settings
Workspace configuration and account management.
Workspace settings
Set the workspace name and image. The image appears in the sidebar so team members can tell workspaces apart. Upload an image or generate one from a color palette.
US or EU AI providers
Workspace admins choose which region’s AI providers the workspace uses. In Workspace Settings, the US and EU AI Models toggle selects between United States, which uses Anthropic and OpenAI, and European Union, which routes all AI requests through Mistral and Black Forest Labs hosted in the EU. Use the EU option when data residency requirements apply. The setting applies to the whole workspace immediately. Members can see the active region in the top-right corner of the app but cannot change it.
API & Developers
The REST API and MCP server let external tools, AI assistants and automations use Growhold’s knowledge, agents and runs.
API keys
Create and manage keys on the API & Devs page. Keys start with gh_ and can be limited to specific areas: knowledge, runs, schedules, agents, specialists. The key value is shown once at creation, so store it then. Revoking a key stops anything using it immediately.
Connecting external tools
The REST API is a single POST endpoint with an action field. All actions, such as agents.list and runs.start, use the same request shape with different payloads. The MCP server exposes the same actions as tools for Claude, Cursor and other MCP-compatible clients. Connect with an API key, or create an OAuth client and add Growhold to Claude.ai as a custom connector. The full endpoint reference is linked below.