Buttercup
Buttercup is a browser-based sandbox for prototyping and testing AI agents without backend infrastructure. It combines weekly interactive lessons on agent architecture with a virtual file system, multi-provider API key management (Anthropic, OpenAI, xAI, Google, Ollama, vLLM), and agent scaffolding from templates. Agencies can debug agents in the browser, preview agent-generated web pages, and export finished code as zip files for client delivery. Built for AI development agencies and custom agent builders, Buttercup eliminates deployment overhead during the development phase but is not a white-label or managed service platform.
Buttercup is an agent builder, integrating with Anthropic, OpenAI, xAI, and Google. InnovaAI scores it 5.1/10 for agency resale.
Agency Audit
Buttercup is a browser-based sandbox for building and testing AI agents through weekly interactive lessons, with no backend infrastructure required. Agencies can scaffold starter agents from templates, test them against multiple model providers (Anthropic, OpenAI, xAI, Google, Ollama, vLLM), and export finished workspaces as zip files for client delivery. It's built for AI development agencies and custom agent builders, but resale potential is limited because Buttercup is primarily an educational tool and development harness, not a white-label client service.
5.1/10
Depends on volume
2d 1-2 days
- Your agency builds custom AI agents for clients and needs a rapid prototyping environment that eliminates backend setup overhead.
- You want to train internal staff on agent architecture through structured weekly lessons before deploying production solutions.
- You need to support multiple model providers (OpenAI, Anthropic, xAI, Google, Ollama) in a single workspace without switching tools.
- You plan to resell Buttercup as a white-label client tool or managed service; the interface displays Buttercup branding and is designed for developers, not end-users.
- Your clients require centralized billing, usage reporting, or audit trails; Buttercup routes API calls directly from the browser with no server-side logging.
- You need production-grade deployment infrastructure; Buttercup is a development and learning harness, not a runtime environment for live agents.
Profit Path
Estimate available after setup inputs
$600–$1.5K/project
Monthly Recurring
From 242 published agency rates in USA, 25th to 75th percentile x 20h of assumed delivery time. Rates are self-reported directory profiles, not observed transactions.
Platform Features
Core capabilities of Buttercup
Weekly interactive lessons with exercises
Buttercup delivers one structured AI agent lesson per week, starting mid-September 2026, covering tool-call loops, tool definitions, context management, browser control, and failure modes. Each lesson includes runnable code and hands-on exercises in the browser sandbox, with a full archive available for missed weeks.
Multi-provider API key management
Manage API keys for Anthropic, OpenAI, xAI, Google, OpenRouter, Ollama, and vLLM in a single workspace. Keys stay in browser localStorage and never touch Buttercup servers, so agencies can test agents against different model providers without reconfiguring infrastructure.
Browser-based agent sandbox with file system
Run AI agent code directly in the browser with a virtual file system for debugging. Read, write, edit, and move files; search with grep and glob patterns; and test agent behavior without deploying to production infrastructure.
Agent scaffolding from templates
Generate starter agents from pre-built templates (e.g., Blocks.AI research agents) and customize them in the workspace. Reduces setup time for common agent patterns like web research, data extraction, or planning workflows.
Preview and interact with agent-built web pages
Render and test web pages generated by agents within the Buttercup interface. Verify agent output and browser control logic before exporting to clients.
Export workspaces as downloadable zip files
Package completed agent code, tool definitions, and configuration as a zip file for client handoff or version control integration. Enables agencies to deliver working agents without requiring clients to access Buttercup.
What Makes Buttercup Different
Unique advantages vs similar tools in this niche
Runs entirely in the browser with no backend
vs Traditional agent development environments requiring server setupThe harness is a static page with a virtual file system and undo stack, working on desktop or phone with no build step.
Weekly structured lessons with hands-on exercises
vs Self-directed learning from scattered documentationEach lesson covers one idea with code and an exercise you run in the tab, building from a bare completion call to a working agent.
Supports local models for privacy
vs Cloud-only AI platformsIt talks to local models like Ollama and vLLM so nothing has to leave your machine.
Latest Updates
Recent releases and improvements for Buttercup
A new agent lesson every week
New2026-09-02buttercup.sh is now a course as well as a harness. One lesson by email every week, a single idea, the code that makes it work, and an exercise you run in the tab. Free, no account, and every exercise runs on a local model if you would rather not spend a cent.
buttercup.sh is open source: an agent harness that lives in your browser
New2026-09-01Eighteen tools, a virtual filesystem and an undo stack in one static page, on desktop or phone, with no backend and no build step. Free forever, and it talks to local models so nothing has to leave your machine.
Value Equation
Outcome-likelihood-time-effort assessment for Buttercup
Value math requires real pricing
The Value Equation (dream outcome × likelihood ÷ time × effort) feeds directly into ROI math. Buttercup has no published pricing, so we hold this section until real numbers are available.
Contact ButtercupPricing
Platform cost for Buttercup
Custom pricing
Buttercup uses custom/enterprise pricing: rates aren't published publicly. Contact their team directly for a quote.
Contact ButtercupMarket Intelligence
Offer + scale economics for Buttercup
Offer economics require real pricing
Offer economics, scale projections, and margin potential all depend on Buttercup's actual platform cost. Once pricing is published or shared with your agency, we'll compute the full breakdown here.
Contact ButtercupInvestment Decision Framework
Strategic vetting analysis for Buttercup
Consider
Favorable fit, worth a closer look
Buy If
4Your agency builds custom AI agents for clients and needs a rapid prototyping environment that eliminates backend setup overhead.
You want to train internal staff on agent architecture through structured weekly lessons before deploying production solutions.
You need to support multiple model providers (OpenAI, Anthropic, xAI, Google, Ollama) in a single workspace without switching tools.
You're building proof-of-concepts for clients and need to export working agent code as downloadable zip files for handoff.
Skip If
4You plan to resell Buttercup as a white-label client tool or managed service; the interface displays Buttercup branding and is designed for developers, not end-users.
Your clients require centralized billing, usage reporting, or audit trails; Buttercup routes API calls directly from the browser with no server-side logging.
You need production-grade deployment infrastructure; Buttercup is a development and learning harness, not a runtime environment for live agents.
Your team lacks JavaScript proficiency; tool definitions and agent logic are written as plain JS functions in the workspace, requiring code literacy.
Bottom Line
Buttercup is a browser-based sandbox for building and testing AI agents through weekly interactive lessons, with no backend infrastructure required. Agencies can scaffold starter agents from templates, test them against multiple model providers (Anthropic, OpenAI, xAI, Google, Ollama, vLLM), and export finished workspaces as zip files for client delivery. It's built for AI development agencies and custom agent builders, but resale potential is limited because Buttercup is primarily an educational tool and development harness, not a white-label client service.
Reality Check
Buttercup stores API keys in browser localStorage and routes model traffic directly to the vendor you select, creating a dependency on client-side key management and no centralized billing or usage tracking for resale. Agencies cannot white-label the interface or offer it as a managed service to non-technical clients.
Low effort: self-service setup with guided onboarding
Academy for Buttercup
Work through it in order: the course for this service first, then the modules behind it.
No Academy modules are published for this service yet. Browse the full Academy
Why this category matters
The commercial case before the tooling.
Core concepts
The mental model you need to price and scope the work.
- Agent Surface OwnershipConcept
Agent Surface Ownership is the principle that the durable asset in an agent deployment is not the builder shell but the layer the agency controls: the client workflow definition, the memory and context store, the tool permissions, and the review checkpoints. Two agencies can configure the same visual builder and ship near-identical agents, which is why shell choice alone rarely defends a retainer. What defends it is owning the surface the agent operates on. Forrester's September 2026 argument that private AI deployments outperform public ones for B2B marketing makes the point commercially: shared model access erases differentiation, so the agency that owns client-specific context and governance keeps the account. Concretely, an agency using Chipp for a white-label client assistant should still own the knowledge sources, action permissions, and escalation rules, because those are what the client cannot replicate by switching vendors. Audit every agent deployment by asking who holds the workflow map, the memory, and the approval gates.
- Governance Surface RatioConcept
Governance Surface Ratio is the relationship between how many agents an agency deploys and how much review, logging, and rollback infrastructure each one demands. Every agent added to a client workflow expands the surface area that must be audited: memory stores, tool permissions, channel access, and failure paths. The ratio matters because agencies price retainers on delivery hours, not on the governance hours that scale with agent count. A single client-facing agent touching CRM data may need one review checkpoint; ten agents across five accounts can require a dedicated ops function. Forrester's September 2026 research found 83% of B2C marketing decision makers already work with AI agents, meaning the governance burden is now a baseline cost, not a differentiator. Agencies that map governance surface before deployment, rather than after an incident, protect both margin and client trust.
- Orchestration Depth LadderConcept
Orchestration Depth Ladder ranks agent-builder platforms by how much of the client workflow the agency actually owns: prompt shell, tool-call routing, memory and state, multi-step orchestration, and finally governance and testing. Most agencies buy at the bottom rung and quote the top rung. The gap is where margin leaks, because a branded chatbot built on Chipp or FormWise is replaceable in a week, while the integration, audit trail, and evaluation harness around it is not. Forrester's September 2026 finding that private deployments outperform shared public models for B2B marketing makes the point commercially: differentiation lives in owned context and controls, not the model call. Climb one rung per quarter against a named client workflow, and price the retainer against the rung you can defend, not the demo you can show.
Decision and risk
How to judge the fit, and the ways it goes wrong.
- Agent Builders Rule: Price the Shell Only After the Client Workflow Has a Named OwnerEvaluation Rule
Name the client workflow, its human owner, and its failure cost first; only then pick the builder whose white-label depth, memory model, and audit surface match that answer.
- Agent Builders Rule: Score the Handoff Before You Score the BuilderEvaluation Rule
Choose the builder whose review, versioning, and rollback path your least technical delivery lead can operate alone, then negotiate the commercial model around that constraint.
- Agent Builders Decision: White-Label Resale Shell vs Governed Internal Delivery LayerDecision Framework
IF a client workflow is repeatable, low-risk, and the agency intends to sell it as a branded product or retainer line, THEN a white-label builder shell (Chipp, FormWise) shortens time-to-revenue because branding, domains, and client seats are already handled. IF the workflow touches client CRM data, outbound communications, or regulated records, THEN the durable choice is a governed internal delivery layer where behavior is versioned, tested, and auditable before any client sees it. The decision is not which builder is better; it is whether the agency is monetizing a shell or owning the controls around it.
- The Demo-to-Delivery Gap: Why Agent Builders Stall After the First Client PilotFailure Pattern
- The White-Label Shell Trap: Why Agent Builders Collapse When the Client Asks for GovernanceFailure Pattern
Delivery system
Blueprints and procedures for running it as a service.
- White-Label Agent Productization Sprint (10-14 days)Implementation Blueprint
A fixed-scope engagement that turns one named client workflow into a branded, governed agent the agency can bill against a retainer instead of reselling a vendor seat. The sprint ships the agent, the controls around it, and the commercial wrapper the agency owns.
- Agent Scope Contract (Onboarding)Operating Procedure
- Agent Builders: Autonomy Boundary Review (QA)Operating Procedure
- Agent Builders: Client Handoff Playbook (Handoff)Operating Procedure
13 modules selected for Buttercup
Frequently Asked Questions
Answers about pricing, setup, implementation
Buttercup is a browser-based development harness for building AI agents through weekly interactive lessons. It provides a sandbox to scaffold agents from templates, test them against multiple model providers (Anthropic, OpenAI, xAI, Google, Ollama, vLLM), debug with a virtual file system, and export finished agent code as zip files. No backend infrastructure or signup is required; API keys stay in your browser.
Buttercup does not publish pricing on its website. Access to the platform and weekly lessons appears to be free during the initial launch phase (starting mid-September 2026). Contact the team via GitHub for current pricing or enterprise licensing details.
No verified white-label program exists. The Buttercup interface displays the Buttercup brand and is designed as a developer tool, not a client-facing service. Agencies can export agent code as zip files for client delivery, but cannot rebrand the Buttercup workspace itself.
Yes. Buttercup supports native integration with Anthropic, OpenAI, xAI, Google, OpenRouter, Ollama, and vLLM. You paste API keys directly into the workspace, and model traffic routes from your browser to the provider you select. No server-side proxy or additional configuration is required.
Initial setup takes under 5 minutes once you paste an API key into the workspace. Scaffolding a starter agent from a template (e.g., a Blocks.AI research agent) takes seconds. Total time to a working prototype depends on agent complexity, but the sandbox eliminates backend infrastructure setup entirely.
Buttercup is designed for AI development agencies, custom agent builders, and technical training providers. It's best suited for clients who need bespoke AI agents (research automation, data extraction, workflow orchestration) and can work with exported code, not for non-technical end-users seeking a managed service.
Yes. The weekly interactive lessons are structured to teach agent architecture from first principles, covering tool-call loops, context management, planning, and failure modes. Each lesson includes runnable code and exercises, making it suitable for onboarding developers to agent-building patterns.
API keys are stored only in your browser's localStorage and never sent to Buttercup servers. When you export a workspace as a zip file, you receive the agent code, tool definitions, and configuration files. Keys are not included in the export; clients must supply their own API credentials to run the agent.