Get Atlas running in 90 seconds.
$ cargo install atlas-cli
Or download a pre-built binary from the GitHub releases page.
$ pip install atlas-sdk
Create a markdown file with a YAML frontmatter block that defines a knowledge package:
---
kind: Package
id: package:my_package
name: My Package
version: 0.1.0
---
# My Knowledge
A **Concept** describes a thing or idea.
Node: concept:my_idea
- Name: My Idea
- Description: This is an idea worth knowing.
An **API** describes a function or method signature.
Node: api:do_thing
- Name: do_thing()
- Signature: do_thing(arg1: string): Result
- Description: Does the thing you need.
Compile your knowledge package into a portable .atlas binary:
$ atlas compile my_package.md --out my_package.atlas
$ atlas solve --bundle my_package.atlas "my idea"
$ python -c "
import atlas_sdk as atlas
result = atlas.solve('my_package.atlas', 'my idea')
for node in result.nodes:
print(node.name, node.kind)
"
Add a decision tree to your package via frontmatter:
---
kind: Package
id: package:my_package
decision_tree:
id: tree:hello
trigger: { tags: [hello] }
root: start
nodes:
- id: start
question: "Say hello?"
branches:
- condition: "yes=true"
next: greet
- id: greet
terminal:
recommendation:
- node_id: concept:my_idea
confidence: 1.0
rationale: "Greet the user."
---
$ atlas decide --bundle my_package.atlas "hello" -c "yes=true"
Every bundle is verifiable. Ensure all edges reference valid nodes:
$ atlas verify --bundle my_package.atlas
Search, fetch, and compile packages from the public registry:
$ atlas search flutter
$ atlas install flutter_core.md
$ python -c "
from atlas_sdk import Agent
agent = Agent('my-agent', packages=['flutter_core'])
result = agent.solve('stateless widget')
print(result.nodes[0].name)
"
Read the API Reference for all CLI commands and Python SDK functions.