Welcome to the GRACE ERA (Editor for Research Acceleration) by GraceKits.com, a powerful, open-source tool designed to serve as your "factory-maker" for ideas. Whether you're a researcher building a scientific theory, a developer creating a software package, or a creator generating content, ERA provides the architecture to structure, automate, and scale your work with AI.
This is more than a prompt editor; it's a complete Integrated Development Environment (IDE) for thought. It allows you to deconstruct complex ideas into reusable components, orchestrate them into powerful workflows, and collaborate with a panel of AI models in a structured, verifiable way.
This guide will walk you through nearly every feature, from basic setup to advanced, multi-step workflows. No prior coding knowledge is required (but would certainly be helpful!), only a desire to build and the time to do so.
1. Introduction: The Factory-Maker Philosophy
The GRACE ERA tool is built on a simple but powerful idea: The gap of time created by the lengthy process of research and "proof" can be accelerated greatly, through prompt engineering and structured data producing tools. In terms of importance, this matters, because it can save lives, it can heal hearts, and possibly offer developing faster solutions for humanity's greatest problems.
Create Reusable Components: Break down complex knowledge and/or text into modular Parts that are reusable and will save you time when using AI for similar or repeated tasks.
Design Individual Workflows: Arrange Parts into logical Workflows called templates.
Build Automated Assembly Lines: Connect and Sequence Workflows and/or web requests into powerful Chains that can accomplish a complex task in a logical order.
Package Your Work: Export your entire "factory" as a single "Grace Kit" (.json file) that you can save, share, and collaborate on with others.
This tool is designed to support a Tri-Unity Framework of thinking, where the structured logic of ERA (the "Left Brain") can be combined with the creative power of external AI models (the "Right Brain"), all guided by human oversight (the "Oracle").
2. Getting Started: Your First Session
ERA is a single HTML file. No installation is needed.
Save the file as GraceERA.html.
Open it in a modern web browser like Chrome, Firefox, or Edge.
Upon first launch, you'll be prompted to set up your session.
Enter Your API Key: To use the AI features, you need a Google AI API Key. You can get one from Google AI Studio. Paste this key into the "Google AI API Key" field.
Security Note: Your key is ephemeral. It's held only in your browser's memory for the current session and is never saved or transmitted to our servers. It is gone when you close the tab.
Select a Model: Choose a Gemini model from the dropdown. The list is updated with the latest versions.
Models (gemini-1.5-flash, gemini-1.5-pro, gemini-2.0-flash, gemini-2.5-flash, gemini-2.5-pro): "Flash" models are best for speed and cost-efficiency. Ideal for structuring, formatting, and simpler tasks.
Pro Models (gemini-1.5-pro, gemini-2.5-pro): Best for power and complex reasoning. Ideal for generating novel theories or detailed manuscripts.
Start Session: Click the button to begin. You're now ready to build.
3. Core Concepts: The Building Blocks of GRACE
Project: A self-contained workspace for a specific goal (e.g., "ALS Research," "Blog Post Content", "Automate Business Task"). Each project has its own library of Parts, Templates, and Chains.
Part: A reusable snippet of text. This can be a piece of evidence, a style guide, a meta-instruction for the AI, or a complex prompt.
Template: A blueprint that combines one or more Parts. This can be used to define the structure and logic of your prompt.
Chain: A multi-step workflow where the output of one step becomes the input for the next. This is your automated assembly line.
Grace Kit: An exported .json file containing a complete, shareable ERA project. It's a "forkable" snapshot of your entire factory.
4. The Workspace: Managing Projects
Create: Click the + button to create a new, blank project.
Switch: Use the dropdown menu to switch between your projects.
Delete: Click the Delete button to permanently remove the currently selected project (you may not delete the default project).
Manage: Click the Manage button to see a list of all Parts, Templates, and Chains in the current project, with options to delete them individually. You will also be warned/notified if you are removing a part for another process is dependent upon. For example, if a template uses a part, and you try to delete the part, you will be asked if you are sure and notified about the dependency.
5. The Raw Materials: Creating & Using Parts
Parts are the atoms of your workflow. They allow you to deconstruct knowledge into manageable, reusable pieces.
Click Create Part.
Name: Give it a unique, descriptive name (e.g., 001_evidence_neurogenesis_study).
Content: Paste the text content. This can be a single sentence or many pages long.
Import from File: Use this to load content directly from a .txt or .md file (or other text file). please note, that if you plan to edit the content in the application, importing will over
Example Part: A Meta-Instruction
Name:meta_instruction_be_verbose Content:System Note: Be as verbose and detailed as possible. Do not worry about the length of your response. You will be prompted to continue if you reach the token limit. Your primary goal is to be thorough, detailed, and comprehensive.
6. The Blueprints: Working with Templates
Templates give your Parts structure and logical flow. A template is a saved collection of Parts.
Click Select Parts and check the boxes for the Parts you want to include in your template.
Click Save Template and give it a name (e.g., biorxiv_manuscript_scaffold).
The tool combines the content of the selected parts in alphabetical order of their names (or Numerical order from lowest to greatest). This is important for structuring your prompts predictably and understanding how to get the response and result you desire using the parts you created and utilize.
7. The Assembly Line: Advanced Workflows with Chains
This is the most powerful feature of ERA. Chains automate complex, multi-step tasks, turning your Parts and Templates into a functioning "factory."
7.1 What Are Chains?
A chain is a sequence of steps where the final output of one step becomes the initial input for the next. There are three types of steps:
AI Step: Executes a prompt using a selected Template and Model.
GET Action: Fetches data from a URL. The input is appended to the URL.
POST Action: Sends data to a URL. The input is sent as a JSON payload.
7.2 Creating a Chain
Click Chains, then Create New Chain.
Give your chain a descriptive name (e.g., wiki_to_explainer_pipeline).
Use the "Add New Step" section to build your workflow. You can reorder steps with the arrows or delete them.
Example Chain: "Topic to Draft"
Step 1: GET Action URL:https://en.wikipedia.org/w/api.php?action=query&prop=revisions&rvprop=content&format=json&origin=*&titles= This step takes a topic name (e.g., "Parkinson's Disease") and fetches the raw Wikipedia article data.
Step 2: AI Step Template:template_clean_and_summarize Model:gemini-2.5-flash Mode:Editor This step takes the raw JSON from Step 1 and uses a template to extract and summarize the key information.
7.3 Understanding AI Step Modes: Editor vs. Build
The mode of an AI Step is a crucial strategic choice that dictates how the AI processes information.
Editor Mode (New Prompt First): The AI receives the instructions from your Template *before* it sees the input from the previous step. This is best for analytical tasks where you want the AI to act *upon* the content.
Use Cases: Summarizing, reformatting, fact-checking, extracting data.
Analogy: Handing an editor a style guide and then giving them a manuscript to edit. `(Instructions -> Content)`
Build Mode (Content First): The AI receives the input from the previous step *before* it sees your Template's instructions. This is best for creative and synthesis tasks where you want the AI to fully absorb the context before it knows what to do.
Use Cases: Writing a new article based on research notes, generating a novel theory from data, creating a new piece of code from a design spec.
Analogy: Giving an architect all the raw survey data and material specs and *then* asking them to design a building. `(Content -> Instructions)`
7.4 Sticky Parts in Chains
When selecting Parts for a Template, you can mark some as **Sticky**. A sticky part's content will be included in *every subsequent AI step* in a chain, not just the one where its template is used.
Use Case: Create a sticky part named persona_master_researcher with the content "You are a world-class research scientist with a talent for clear communication." If used in the first template of a chain, every AI in that chain will adopt that persona.
8. Executing Your Work: Go, Run, & Loop
Go: Executes a simple, one-time prompt using the text in the prompt box combined with any currently selected Parts. Note, the Go function uses the default AI model you selected at login / API key entry.
Run Chain: Opens a modal to select and execute one of your saved Chains. Keep in mind, that when you built the chain and assigned templates, you were able to select the AI model individually for each template / "chain step".
Looping a Chain: In the "Run Workflow Chain" modal, you'll find a Loop Chain checkbox and a delay field.
How it works: When checked, the chain will complete a full run. Then, after the delay, it will start over from Step 1, using the final output of the previous run as the new input.
Use Case: Perfect for creating autonomous agents that work through a task list, iterative refinement processes where a document is improved with each pass, or continuous monitoring of an API endpoint.
Stopping the Loop: The main Cancel button on the interface will immediately stop any active process *and* terminate the loop.
9. Utilities & Data Management
Preview: See the exact, final prompt that will be sent to the AI before you run it. Indispensable for debugging complex templates.
Build Megaprompt: A convenience function to combine all selected Parts into a single .txt file for use in other applications BUT DOES NOT include anything typed in the prompt window.
Import/Export: Your project's lifeline.
Export: Saves your entire workspace (all projects, parts, templates, and chains) to a single .json file. This is your **Grace Kit**.
Import: Loads a Grace Kit .json file, completely restoring a saved workspace. This is the primary mechanism for collaboration and backing up your work.
10. Advanced Use Cases: The Path to Autonomy
The true power of ERA lies in combining these features to create autonomous workflows.
The Oracle Network
NOTE: This section requires some technical experience. The script needs to be connecting to a page that allows CORS (cross-origin) and you cannot simply "scrape" or "capture" a pages content. Use a `POST Action` to call a custom PHP script that acts as a "router." This script can call multiple external AIs (Grok, GPT-4o, etc.) to get "second opinions" or creative feedback, then return the synthesized result to your ERA chain for the next step. This creates a panel of experts for your agent to consult.
The "Expand then Contract" Workflow
A powerful two-chain process for generating high-quality, factual content:
Chain 1 (Expand): Use a `Build Mode` AI step with a `meta_instruction_be_verbose` part to generate a massive, detailed, creative draft on a topic from a source document. For example, you would encourage the AI to be verbose be very detailed, unabridged, and comprehensive in every way.
Chain 2 (Contract): This chain takes two inputs: the verbose draft and the original source document. An `Editor Mode` AI step then uses a template with the instruction: "Fact-check the draft against the source of truth and modify the draft in order to make it fully align with the source of truth provided, which is the original source document." This uses the AI to eliminate its own hallucinations, resulting in a highly accurate final document.
The Self-Terminating Loop
Design a loop where the final step is a `POST Action` to a PHP script that analyzes the output. If the output meets a certain quality standard, the script could be instructed to return a `404 error`. This intentionally causes the ERA chain to fail and break the loop, signaling a successful completion of the task.
11. Best Practices and Tips
Think Atomically: The more you break down your knowledge into small, single-purpose Parts, the more flexible and powerful your system will be.
Name Everything Clearly: Good naming conventions for Parts, Templates, and Chains (e.g., 001_part_evidence_..., template_format_..., chain_research_...) are crucial for managing complex projects.
Build Incrementally: Test each step of a chain individually before connecting them. Test a chain with a single run before attempting a loop.
Use the `Preview` Button Liberally: Always check your final prompt before a complex `Go` or `Run Chain` to ensure it's structured as you expect.
12. Troubleshooting
"Failed to fetch" Error: Almost always a CORS issue on a GET/POST action or a malformed URL. The target server must be configured to allow requests from other domains. For public APIs, look for an &origin=* parameter or similar to include in your URL.
API Key Error: Double-check that your key is correct, has been enabled in your Google AI account, and does not have any IP restrictions that would block it.
Chain Fails on a Step: Isolate that step. Check that the template name is correct, the URL is valid, or the input data is in the expected format.
13. Glossary
Build Mode
An AI step setting where the input from the previous step is placed before the template's content. Ideal for synthesis and creative tasks where context is paramount.
Chain
A sequence of steps (AI calls, GET/POST actions) where the output of one step serves as the input for the next, enabling automated workflows.
CORS (Cross-Origin Resource Sharing)
A web security mechanism that controls whether a script on one website is allowed to request resources from another. A common cause of "Failed to fetch" errors if not configured correctly on the target server.
Editor Mode
An AI step setting where the template's content is placed before the input from the previous step. Ideal for analytical and formatting tasks.
Ephemeral
Lasting for a very short time, not persistent or static. In ERA, your API key is ephemeral, meaning it exists only in your browser's active memory and is gone when you close the tab.
Grace Kit
An exported .json file containing a complete, shareable ERA project. It is the "source code" of a knowledge factory, allowing for backup, sharing, and collaboration ("forking").
Oracle
An external API, often another AI model, that a chain calls via a GET or POST action to retrieve information, get a second opinion, or perform a specialized task that the primary AI cannot.
Part
An atomic, reusable snippet of text that serves as a building block for prompts. Parts can contain instructions, evidence, context, or any other piece of information.
Sticky Part
A special type of Part that, once included in a template, will have its content prepended to *every subsequent AI step* within the same chain run, providing persistent context or persona.
Template
A saved collection of Parts that defines the logical structure of a prompt acting as a workflow.