Getting Started
Get from a clean machine to a validated, running agent without an API key, wallet, or live funds.
Want the shortest possible path? Use First 10 Minutes. Need help choosing an install or feature guide? Use Choose Your Path.
Install
Note: Aether Forge is not yet published to PyPI. Until then, install directly from GitHub.
python3 -m venv .venv
source .venv/bin/activate
pip install 'aether-forge @ git+https://github.com/HeyElsa/aether-forge.git'For local framework development:
git clone https://github.com/HeyElsa/aether-forge.git
cd aether-forge
pip install -e '.[dev]'Requires Python 3.12+. The core install has one runtime dependency: jsonschema.
Verify the Runtime
forge doctorExpected shape:
[ ok] Python version: Python 3.12+
[ ok] jsonschema: jsonschema ...
[ ok] Memory store (SQLite): Layer 3 round-trip ok (write + read)
Healthy (with optional skips) — ...Skipped wallet, knowledge, or security checks are fine for the first run. They are optional extras.
Generate a Zero-Risk Agent
This first agent uses the offline heuristic planner. It does not call an LLM, create a wallet, move money, or require provider credentials.
forge generate-fast \
--name "hello-agent" \
--idea "read an input, reason about it, and report a short summary" \
--output ./hello-agent \
--planner-mode heuristicThis creates a complete agent directory:
hello-agent/
aether-forge.json
agent-spec.json
capability-manifest.json
policy-bundle.json
scenario-pack.json
scaffold.manifest.json
Dockerfile
Makefile
tests/test_agent.pyValidate and Evaluate
forge validate ./hello-agent
forge eval-pack ./hello-agentThe generated agent also ships day-one developer targets:
cd ./hello-agent
make test
make eval-pack
make doctorRun One Tick
forge run ./hello-agent \
--max-ticks 1 \
--interval 0 \
--environment sandbox \
--auto-approveYou should see one completed tick and a short summary. That is the minimal Aether Forge loop:
Planner -> Policy Gate -> Execute -> Step Ledger -> MemoryAdd an LLM Planner
When you want real LLM planning, set a provider key and regenerate or override at run time.
export ANTHROPIC_API_KEY=sk-ant-...
forge generate-fast \
--name "research-agent" \
--idea "summarize trusted sources into a daily brief" \
--output ./research-agent \
--planner-mode anthropic \
--planner-model claude-sonnet-4-5 \
--planner-api-key-env ANTHROPIC_API_KEYIf you omit --planner-mode, Aether Forge auto-detects cloud keys first, then local Ollama when no cloud key is set, then heuristic fallback. The resolved planner is saved in aether-forge.json so another machine can run the same agent with the same environment variables.
Add Crypto Features
Crypto is an optional module layer, not required by the core framework.
pip install 'aether-forge[wallet] @ git+https://github.com/HeyElsa/aether-forge.git'
forge generate-fast \
--name "eth-watcher" \
--idea "monitor ETH market conditions and write observations" \
--output ./eth-watcher \
--wallet \
--planner-mode heuristicFor crypto scaffolds, --environment controls policy and promotion context while --mode controls the trading backend:
# Real market-style path with simulated orders
forge run ./eth-watcher --environment sandbox --mode paper --auto-approve
# Live capital path, only after promotion evidence and funding
forge run ./eth-watcher --environment production --mode liveUse from TypeScript
The companion @aether-forge/sdk package validates artifacts and parses planner output from Node, browsers, and edge runtimes. It is intentionally interface-only in v0.1.x; the runtime tick loop remains Python-side.
npm install @aether-forge/sdkimport { assertValid, parsePlannerOutput, validateAgentSpec } from "@aether-forge/sdk";
const spec = assertValid(validateAgentSpec(jsonFromDisk));
const plan = parsePlannerOutput(rawLlmResponse);Next Steps
- First 10 Minutes — shortest safe path from shell to one completed tick
- Choose Your Path — pick the right guide for your goal
- Install Matrix — install only the extras your agent needs
- Generated Agent Anatomy — understand every file
forge generate-fastcreates - End-to-End Tutorial — build and inspect a full generated agent
- Production Readiness — staging, policy, evidence, secrets, and incident controls
- Multi-Tenant Integration — embed agents in a SaaS app
- Extending the Framework — custom planners, data sources, memory stores, and plugins
- CLI Reference — every
forgecommand - Troubleshooting — common failures and fixes