Load your writing. Guide Ailly to your voice. Write your outline. Prompt Ailly to continue to continue the writing. Edit its output, and get even more like that.
Rhymes with Daily.
Ailly's best feature is rapidly itearting on prompt engineering. By keeping your prompts in snippets on the file system, you can make very fine-grained changes to your prompt and immediately see the difference. You can also use all your normal source control tooling to track changes over time - both your changes, and what the LLM does.
- Create a folder,
content. - Create a file,
content/.aillyrc, and put your top-level prompt instructions.- Include system prompts, level setting expectations. etc.
- Create several files,
content/01_big_point.md,content/02_second_point.mdetc. - Run ailly using NodeJS:
npx @ailly/cli --root content
These properties can be set in a combination of places, includeing the command line, .aillyrc, and greymatter. Later settings override earlier.
combinedbooleanIf true, the file's body is the response and the prompt is in the greymatter keyprompt. If false, the file's body is the prompt and the response is in{file_name}.ailly.md. Default false.skipbooleanIf true, the prompt will not be sent through the LLM (but it will be part of the context).isolatedbooleanIf true, the LLM inference will only include the system prompt, and not the prior context in this folder.contextconversation|folder|noneconversation(default) loads files from the root folder and includes them alphabetically, chatbot history style, before the current file when generating.folderincludes all files in the folder at the same level as the current file when generating.noneincludes no additional content (including no system context) when generating.- (note: context is separate from isolated. isolated: true with either 'content' or 'folder' will result in the same behavior with either. With 'none', Ailly will send only the prompt when generating.)
PLAN to use Ailly effectively. Iterate often. Provide context. Put words in Ailly's mouth.
- Prepare a precise prompt (by writing an aillyrc system prompt, providing supporting documents, and giving individual prompt steps).
- Leverage LLM models (by running Ailly on some or all parts of the context chain).
- Assess the generated content (as Ailly and the LLM writes output, make sure it's on the right track).
- Narrow your context (by editing Ailly's generated content to keep the conversation going where you want it to).
- OpenAI
openai - Bedrock
bedrock- Claude 2 and available models documented by AWS
- Enable models in Bedrock console. Remember that models are enabled per-region - you will need to enable them for each reach you call Bedrock from.
- Each model behaves lightly differently, so the default formatter might not work. In that case, either open an issue or poke around in
./core/src/engine/bedrock.
To choose an engine, export AILLY_ENGINE=[bedrock|openai] or provide ailly --engine on the command line.
- Clone the repo and install dependencies
git clone https://github.com/davidsouther/ailly.git ; cd ailly ; npm install
- Compile the core module with
npx tsc -p core- Rerun this step for any edits in
core
- Rerun this step for any edits in
- Install ailly cli with
npm install -g ./cli - Set any environment variables for your engine
export OPENAI_API_KEY=sk-...export AILLY_ENGINE=bedrockdefault: openai, others depending on version.
- Run ailly with
npx aillycd content/33_dad_jokesnpx ailly .
- Optionally, create an alias to run ailly
- Directly with
alias ailly="$(PWD)/cli/index.js - For zsh:
echo "alias ailly='$(PWD)/cli/index.js'" >> ~/.zshrc - For bash:
echo "alias ailly='$(PWD)/cli/index.js'" >> ~/.bashrc - General *nix:
echo "alias ailly='$(PWD)/cli/index.js'" >> ~/.profile
- Directly with
This is powered by Next.js using App Router.
- Clone the repo, install dependencies, and duplicate the env file for local keys.
git clone https://github.com/davidsouther/ailly.git ; cd ailly ; npm install ; cp .env .env.local
- Update
.env.localwith your OpenAI API key. - Clear out the
content/folder, and replace it with your writing.- Open the
contentfolder and see the example system and prompt files. - Conversations are broken into two files,
<nn>p_<name>.mdand<nn>r_<name>.md. - The
<nn>p_<name>.mdis given to the AI, and the response is written to<nn>r_<name>.md - Files in the same folder are part of a single conversation, and kept in order by sorting based on the numeric values of
<nn>. - The
<name>is for you to keep a file name in mind. - See Specification for further details on how the
contentfolder is organized. - TODO: Provide content importers for email
- TODO: Provide instructions for "best practices" creating
- Open the
- Start the project locally with
npm run dev --workspace packages/web- Default at http://localhost:3000
- Visit the
/contentroute.- Generate all prompts with the "Generate all" button.
- The token count is an estimate and the pricing is advisory; there are no limits placed on token size during the API request.
- Generate all prompts with the "Generate all" button.
- Prepare the project for deployment with
npm run export - Good luck!
- Clone the repo and install dependencies
git clone https://github.com/davidsouther/ailly.git ; cd ailly ; npm install
- Package the extension with
npm run package - In VSCode extensions, install
./extension/ailly-0.0.1.vsixfrom vsix. - Right click a file in content explorer and select
Ailly: Generate
- Run
npx tsc -w -p core - Start the
Run Ailly Extensiontask- Choose
tsx: watch - extension/tsconfig.json
- Choose
- When the new window appears, open a folder with your content.
- Ailly currently only works on the first folder in a workspace.
- In debug mode, Ailly disables other extensions and runs in a clean profile. Comment out the
"--profile-temp",line inlaunch.jsonto use your current VSCode profile instead. - WARNING: Ailly may prompt for your OpenAI API Key. If it does so, it will store the key in workspace settings for later access. Do not commit this key to source control!
- If you don't like that behavior, ensure you have
OPENAI_API_KEYset in your environment for VSCode. - Or send a PR with a better way to load the key safely.
- Right Click a file or folder -> Ailly: Generate
- Open a file -> Cmd+P -> Ailly: Generate
Ailly can use plugins to provide additional data when calling the LLM models (retrieval augmented generation, or RAG).
The default plugin rag (AILLY_PLUGIN=rag, --plugin=rag) uses a vectra database in ./vectors/index.json.
You can provide custom plugins with --plugin=file:/absolute/path/to/plugin.mjs.
This plugin must export a default factory function matching the PluginBuilder interface in core/src/plugin/index.ts.
Create an outline of a larger document or written work.
Create a folder for the content, say campaign.
In this folder, create a file .aillyrc, with a high level description of the work.
This will be used as part of the system prompt for every request, so typical prompt engineering practices work well.
For instance, "You are a fantasy tabletop roleplaying game master. The book is a campaign for a high-magic, high-fantasy setting."
Within this folder, create additional folders for each chapter sized section of the outline.
Within each of those folders, create another .aillyrc file with a description of that section.
Continuing the RPG campaign example, this could be "Chapter 1 describes the overall setting. This world is called MyLandia. It is one large continent with two political factions, a republic and an empire."
Then, create one file for each section.
Each section will be an individually generated response from the LLM.
Files will be generated in their alphabetical order, so files (and folders) should generally use two-digit alphanumeric prefixes to determine their sort order.
For typical writing projects, this will be a file in markdown syntax with an optional YAML frontmatter block.
Add YAML frontmatter with a property, prompt, which will be the human prompt to the LLM.
10_republic.md:
---
prompt: >
The Republic is an aristocratic society founded on a shifting structure of familial wealth.
Families elect one member to the ruling senate.
The senate elects a consol for 18 month terms.
Describe more about the Republic's politics and society.
---
20_empire.md:
---
prompt: >
The Empire is a strict military hierarchy across their entire society.
Denizens are not required to join the military, but that is the only path to citizenship.
Leadership is meritocratic, with strictly defined qualification criteria for promotion.
Describe more about the Empire's politics and society.
---
After creating several chapter folders and section files, your content is ready for a first round of Ailly.
From the command line, cd into the content folder. Then, run ailly:
$ npx ailly
You should see output similar to this:
Loading content from /.../70_mylandia
Loading content from /.../70_mylandia/10_setting
Found 2 at or below /.../70_mylandia/10_setting
Found 2 at or below /.../70_mylandia
Generating...
Ready to generate 1 messages
Running thread for sequence of 2 prompts
Calling openai [
{
role: 'system',
content: 'You are a fantasy tabletop roleplaying game master...',
tokens: undefined
},
{
role: 'user',
content: 'The Republic is an aristocratic society founded on...',
tokens: undefined
}
]
...
Review the generated files. Make any edits you want. Then, if you want to rerun just one file (based on, say, changing its prompt or editing earlier responses), provide that path on the command line.
$ npx ailly 10_setting/20_empire.md
Loading content from /.../70_mylandia
Loading content from /.../70_mylandia/10_setting
Found 2 at or below /.../70_mylandia/10_setting
Found 2 at or below /.../70_mylandia
Generating...
Ready to generate 1 messages
Running thread for sequence of 1 prompts
Calling openai [
{
role: 'system',
content: 'You are a fantasy tabletop roleplaying game master...',
tokens: undefined
},
{
role: 'user',
content: 'The Republic is an aristocratic society founded on...',
tokens: undefined
},
{
role: 'assistant',
content: 'The Republic, known officially as the Resplendent ...',
tokens: undefined
},
{
role: 'user',
content: 'The Empire is a strict military hierarchy across t...',
tokens: undefined
}
]
Response from OpenAI for 20_empire.md { id: 'chatcmpl-1234', finish_reason: 'stop' }
Ailly's --edit flag enables specific functionality to achieve good edits to documents.
It employs several tricks to reduce the model's explanations, focusing only on getting replacement or insertion text. For the Claude family, this includes using the Haiku model by default, as well as limiting responses to the first markdown code fence ```.
ailly --edit requires specifying the lines to edit, either as [start]:[stop] which are 1-based includes/exclusive replacement markers, or as [start]: or :[stop] as 1-based insert after or before, respectively.
Ailly will show the proposed change and ask for confirmation before editing the file, though the prompt can be suppressed by passing --yes on the command line.
Here's an example, using the createUserPoolHandler.js file from content/40_ramdectomy.
% ailly --context folder --prompt "Rewrite the storeUserPoolMeta function to use vanilla javascript without using Ramda." createUserPoolHandler.js --edit --lines 14:19
Edit createUserPoolHandler.js 14:19
import { createUserPoolClient } from "../../../actions/create-user-pool-client.js";
import { join } from "ramda";
-const storeUserPoolMeta = (...args) => {
- const tmp = getTmp(FILE_USER_POOLS);
- const entry = join(",", args);
- setTmp(FILE_USER_POOLS, tmp ? `${tmp}\n${entry}` : entry);
-};
+const storeUserPoolMeta = (...args) => {
+ const tmp = getTmp(FILE_USER_POOLS);
+ const entry = args.join(",");
+ setTmp(FILE_USER_POOLS, tmp ? `${tmp}\n${entry}` : entry);
+};
const validateUserPool = (poolName) => {
if (!poolName) {
Continue? (y/N) y
% git diff .
diff --git a/content/40_ramdectomy/createUserPoolHandler.js b/content/40_ramdectomy/createUserPoolHandler.js
index 5bc1fdc..54fe19f 100644
--- a/content/40_ramdectomy/createUserPoolHandler.js
+++ b/content/40_ramdectomy/createUserPoolHandler.js
@@ -13,7 +13,7 @@ import { join } from "ramda";
const storeUserPoolMeta = (...args) => {
const tmp = getTmp(FILE_USER_POOLS);
- const entry = join(",", args);
+ const entry = args.join(",");
setTmp(FILE_USER_POOLS, tmp ? `${tmp}\n${entry}` : entry);
};
This is a bit of an experimental / trial by doing issue. Generally, late 2024 LLMs (Llama, Claude, ChatGPT) respond with "around" 700 words. Similarly, while their context sizes vary, they seem to work best with "up to" 16,000 words. Based on these limits, up to about 5 layers of folder depth and at most 20 files per folder generate the most reliably "good" results, but your milage will vary.
In LLM Chatbots like ChatGPT or chains like Langchain, the history of the conversation is kept in the sequence of human, assistant interactions. This is typically kept in memory, or at least in an inaccessible format to the user. The user can only regenerate sequences, or add their next prompt at the end.
Ailly removes this limitation by using your file system as the conversational history. The writer maintains full control over the sequence of prompts, up to and including editing the LLM's response before (re)generating the next prompt! This lets the writer decide how the conversation should evolve. By editing an LLM prompt, they can take the best of what the LLM did in some cases, and modify it in others. Using Ailly's filesystem based conversational history, each piece of the session can be stored in source control. Version tracking lets the author see how their prompts and responses have changed over time, and unlock a number of long-term process improvements that are difficult to impossible with chat interfaces.
In one session, a developer was working on a long sequence of prompts to build a software project.
While reviewing an LLM written draft of the README, the developer wanted the list of API calls to be links to the reference documentation.
With a chat conversational history, the developer would have needed to modify the instructions for the entire prompt to encourage creating the list, rerun the generation, and hope the rest of the README came out similarly.
Instead, with Ailly, the developer created a new file with only the list and an instruction on how to create URLs from the list items, saved it as list.md (with isolated: true in the combined head), and ran ailly list.md.
The LLM followed the instructions, generated just the updated list, and the developer copied that list into the original (generated) README.md.
In later prompts, the context window included the entire URLs, and the agent model was able to intelligently request to download their contents.
To the author's knowledge, no other LLM interface provides this level of interaction with LLMs.