Merge pull request kreneskyp#35 from kreneskyp/generic_agents

v0.1 merge

Major features
- langchain chains
- artifact system
- planning system
- multi-agent chat
- real time chat stream
- graphical chain viewer
- refined ux for messages
- refined ux light/dark mode

.gitignore

@@ -9,6 +9,8 @@ node_modules
 .pyc
 .coverage
 __generated__
+frontend/schema.graphql
+.ipython
 
 __pycache__
 
@@ -18,3 +20,6 @@ public
 react-app-unused
 tmp
 workspace
+.bak
+*.bak
+workdir

CHANGELOG.md

@@ -0,0 +1,90 @@
+
+
+## 0.1 - Alpha release of core functionality.
+
+This is the first tagged release and it focuses on establishing some core functionality. This project is
+still in the early stages of development and is not ready for general use. It is in a state that allows
+developers to begin hacking chains and agents.
+
+### Langchain
+The `AgentProcess` has been converted to use langchain chains, LLM, and memory. This is introduces support for
+multi-step LLM processes. 
+
+The default agent is now a zero-shot agent that runs a root chain. A `LLMToolChooser` may be used to replicate 
+decision making of an agent. Future releases will introduce additional agent types.
+
+#### JSON Config
+A custom JSON config loading mechanism was introduced for loading langchain chains. This allowed for loading
+custom chain classes provided by Ix and integrating Ix specific functionality. 
+
+It is a goal to provide better interoperability between vanilla langchain config objects and Ix. This will
+be explored in a future release.
+
+#### Data model
+
+New models `ChainNode` and `ChainEdge` were introduced to store langchain chains. This includes helper methods
+to convert to/from `ChainNode` and `ChainEdge`, a json config, and a `Chain` instance.
+
+#### Custom Chain types
+Multiple custom chains were added to support the Ix style of implementing chains and to add flow control.
+
+New basic chains:
+- LLMToolChain: chain that has `tools` available in the prompt.
+- LLMChain: wrapper around `langchain.LLMChain` to add config loader.
+- LLMReply: chain that replies to a message with LLM prompt.
+- ParseJSON: chain that parses a JSON string into a python object.
+
+New routing / flow control chains:
+- IxSequence: wrapper for Sequence that provides config loader
+- MapSubchain: chain that runs a subchain for each value in a list
+- ToolChooser: chain that chooses a tool with a subchain
+- LLMToolChooser: chain that chooses a tool with LLM prompt
+
+New chains were added:
+- ChatModerator: moderates a chat by delegating tasks to other agents.
+- Planner v3: plans and executes a sequence of tasks.
+
+Test / Example chains were added:
+- MockChain: echos inputs back for tests.
+- Fake Weather: generates fake weather data.
+- Dad Jokes: tells dad jokes.
+
+
+### Multi-agent chat
+Chats now include the ability to add multiple agents. The main agent is a moderator that delegates tasks
+to the other agents. Agent's may be targeted directly by starting a message with a @mention.
+
+
+### UX Updates
+
+#### Chain viewer
+This release introduces a chain viewer that allows the user to view a graphical representation of a chain. The
+viewer will be converted to an editor in a future release.
+
+#### Chat
+- Chat replaces the task view as the default view
+- Chat message updates are now over websockets for real-time updates.
+- Chat messages are now grouped by execution to provide a clean and concise view of the chat.
+- Chat messages now indicate the agent that sent the message.
+
+
+#### Misc
+- various tweaks to improve light/dark style 
+
+### Artifacts
+
+This version introduces the artifact system. Artifacts represent the result of tasks the agent has performed.
+It provides a type of object permanence and common understand for the agent. Artifacts are stored in the database
+and can be viewed by the user or used by the agent in future tasks.
+
+Future release will introduce the ability to reference artifacts in tasks and to create new artifacts 
+from existing ones.
+
+### System
+
+- Webserver is now Nginx and uvicorn to support websockets and http requests.
+- Django upgraded to 4.2.1
+- Switched to psycopg v3 for async query support.
+- Django channels is now available for inter-process communication.
+- ReactFlow is now available to frontend
+- Webpack now supports css to support ReactFlow

Makefile

@@ -1,3 +1,4 @@
+
 DOCKER_COMPOSE=docker-compose.yml
 DOCKERFILE=Dockerfile
 DOCKER_REGISTRY=ghcr.io
@@ -7,8 +8,8 @@ IMAGE_TAG=$(shell cat $(HASH_FILES) | md5sum | cut -d ' ' -f 1)
 IMAGE_URL=$(DOCKER_REPOSITORY):$(IMAGE_TAG)
 IMAGE_SENTINEL=.sentinel/image
 
-DOCKER_COMPOSE_RUN=docker-compose run --rm sandbox
-DOCKER_COMPOSE_RUN_WITH_PORT=docker-compose run -p 8000:8000 --rm sandbox
+DOCKER_COMPOSE_RUN=docker-compose run --rm web
+DOCKER_COMPOSE_RUN_WITH_PORT=docker-compose run -p 8000:8000 --rm web
 
 # set to skip build, primarily used by github workflows to skip builds when image is cached
 NO_IMAGE_BUILD?=0
@@ -59,13 +60,13 @@ frontend: compose npm_install graphene_to_graphql compile_relay webpack
 
 # install npm packages
 .PHONY: npm_install
-npm_install: compose
-	docker-compose run --rm sandbox npm install
+npm_install: compose package.json
+	docker-compose run --rm web npm install
 
 # compile javascript
 .PHONY: webpack
 webpack: compose
-	docker-compose run --rm sandbox webpack --progress
+	${DOCKER_COMPOSE_RUN} webpack --progress
 
 # compile javascript in watcher mode
 .PHONY: webpack-watch
@@ -75,19 +76,26 @@ webpack-watch: compose
 # compile graphene graphql classes into schema.graphql for javascript
 .PHONY: graphene_to_graphql
 graphene_to_graphql: compose
-	docker-compose run --rm sandbox ./manage.py graphql_schema --out ./frontend/schema.graphql
+	${DOCKER_COMPOSE_RUN} ./manage.py graphql_schema --out ./frontend/schema.graphql
 
 # compile javascript
 .PHONY: compile_relay
 compile_relay: compose
-	docker-compose run --rm sandbox npm run relay
+	${DOCKER_COMPOSE_RUN} npm run relay
 
 
 # =========================================================
 # Run
 # =========================================================
 
-# run backend and frontend
+# run backend and frontend. This starts uvicorn for asgi+websockers
+# and nginx to serve static files
+.PHONY: server
+server: compose
+	docker-compose up web nginx
+
+
+# run django debug server, backup in case nginx ever breaks
 .PHONY: runserver
 runserver: compose
 	${DOCKER_COMPOSE_RUN_WITH_PORT} ./manage.py runserver 0.0.0.0:8000
@@ -126,7 +134,10 @@ migrations: compose
 # load initial data needed for dev environment
 .PHONY: dev_fixtures
 dev_fixtures: compose
-	${DOCKER_COMPOSE_RUN} ./manage.py loaddata fake_user.json default_agent.json
+	${DOCKER_COMPOSE_RUN} ./manage.py loaddata fake_user
+	# use management commands for now since fixtures didn't load correctly
+	${DOCKER_COMPOSE_RUN} ./manage.py create_moderator_v1
+	${DOCKER_COMPOSE_RUN} ./manage.py create_coder_v1
 
 
 # =========================================================

README.md

@@ -29,9 +29,12 @@ knowledge, casting a shadow of intrigue over the galaxy.
 
 ## About
 <div>
-iX is a platform to run semi-autonomous GPT-4 agents, providing a scalable and responsive solution for delegating tasks.
-Agents can be spawned as individual processes to research and complete tasks through a web based interface while the 
-backend architecture efficiently manages message queues and inter-agent communication between a fleet of workers.
+Ix is an experimental platform for designing and deploying semi-autonomous LLM agents. It provides a scalable and
+responsive solution for delegating tasks to AI powered agents. The platform is designed to be extensible, allowing
+developers to create custom agents and chains to perform a wide variety of tasks.
+
+The backend is designed to support multiple agents running in parallel and communicating with each other. Each agent
+may be customized and may utilize parallel processes to complete tasks.
 <br>
 <br>
 The platform supports deployment using Docker containers, ensuring a consistent environment and enabling easy scaling 
@@ -40,44 +43,41 @@ to multiple worker containers.
 
 ## How does it work
 
-You provide a task and goals and an agent uses that direction to investigate, plan, and complete tasks. The agents are
+### Basic Usage
+You chat with an agent that uses that direction to investigate, plan, and complete tasks. The agents are
 capable of searching the web, writing code, creating images, interacting with other APIs and services. If it can be 
-coded, it's within the realm of possibility for an agent to assist.
+coded, it's within the realm of possibility that an agent can be built to assist you.
 
-1. Start by create a task and goals for the agent to work towards.
-    ![Dialog for entering task name and goals](docs/create_task.png)
+1. Setup the server and visit `http://localhost:8000`, a new chat will be created automatically
 
-2. The agent will begin researching and planning to complete the task.
-    ![chat interface displaying log](docs/chat.png)
+2. Enter a request and the Ix moderator will delegate the task to the agent best suited for the response. Or @mention
+an agent to request a specific agent to complete the task.
 
-3. The chat system provides an interaction model for the agent and user to communicate. The agent will respond to
-    user input and provide updates on its progress. The Agent will request additional information and authorization
-    from the user as needed.
+3. Customized agents may be added or removed from the chat as needed to process your tasks
 
-    ![chat interface displaying log](docs/chat_interactions.png)
 
-4. Agents are customizable and can be extended to support new functionality. The platform provides a plugin architecture
-    for adding new agents and customizing existing agents.
-    ![chat interface displaying log](docs/model_options.png)
+### Creating Custom Agents and Chains
 
+View the [documentation](docs/chains/chains.rst) to create custom agents and chains.
 
 
 ## Key Features
 
 - Scalable model for running a fleet of GPT agents.
-- Responsive user interface. for interacting with agents.
+- Responsive user interface for interacting with agents.
 - Persistent storage of interactions, processes, and metrics.
 - Message queue for agent jobs and inter-agent communication.
 - Extensible model for customizing agents.
 - Deployment using Docker.
 
 
-## Technologies:
+## Stack:
 - Python 3.11
 - Django 4.2
 - PostgreSQL 14.4
-- Graphql
+- Graphql / Graphene / Relay
 - React 18
+- Langchain
 - Integrated with OpenAI GPT models
 - Plugin architecture to support extending agent functionality (e.g. web browsing, writing code, etc)
 - Generic framework for vector database based agent memory
@@ -131,7 +131,7 @@ make dev_setup
 Run the dev server
 
 ```bash
-make runserver
+make server
 ```
 
 Start a worker
@@ -155,7 +155,7 @@ Run as many worker processes as you want with `make worker`.
 Here are some helpful commands for developers to set up and manage the development environment:
 
 ### Running:
-- `make runserver`: Start the application in development mode on `0.0.0.0:8000`.
+- `make server`: Start the application in development mode on `0.0.0.0:8000`.
 - `make worker`: Start an Agent worker.
 
 ### Building:

bin/celery.sh

@@ -2,5 +2,6 @@
 set -o errexit
 
 CELERY_CONCURRENCY=${CELERY_CONCURRENCY:=1}
+DJANGO_SETTINGS_MODULE=ix.server.celery_settings
 
 celery -A ix worker --loglevel=info --concurrency=${CELERY_CONCURRENCY}

docker-compose.yml

@@ -9,8 +9,12 @@ services:
       POSTGRES_DB: ix
       POSTGRES_HOST_AUTH_METHOD: trust
 
-  sandbox:
+  web:
     image: ghcr.io/ix/sandbox:latest
+    #image: ghcr.io/ix:latest
+    command: uvicorn ix.server.asgi:application --host 0.0.0.0 --port 8001 --reload
+    ports:
+      - "8001:8001"
     links:
       - db
       - redis
@@ -19,12 +23,21 @@ services:
       #- .node_modules:/var/npm/node_modules
       - ./bin:/usr/bin/ix
       - .bash_profile:/root/.bash_profile
+      - .ipython:/root/.ipython
     env_file:
       - .env
     environment:
       DJANGO_SETTINGS_MODULE: "ix.server.settings"
+
+  nginx:
+    image: nginx:latest
     ports:
       - "8000:8000"
+    volumes:
+      - ./nginx.conf:/etc/nginx/nginx.conf
+      - ./.compiled-static:/var/static/
+    depends_on:
+      - web
 
   worker:
     image: ghcr.io/ix/sandbox:latest
@@ -41,7 +54,6 @@ services:
       - .env
 
 
-
   redis:
     image: redis