AgentPi is a signal-driven edge intelligence runtime designed for Raspberry Pi and similar edge devices. It converts real-world sensor signals into context-aware agent behaviors using a modular, event-driven architecture.

python -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
python -m pip install -e .
python -m agentpi --duration 10

Note: This repo contains a minimal demo pipeline that runs a mock signal adapter, processes signals into events, updates context, and activates a summarizer agent.

AgentPi follows a signal → observation → event → context → activation → agent pipeline.

1. Signal ingestion (adapters)
2. Observation creation (cognition)
3. Event generation (cognition)
4. Context update (context builder)
5. Activation evaluation (activation engine)
6. Agent responses (agents)
7. Persistence (memory)

• agentpi/ — core Python package
  • adapters/ — signal producers (hardware, mock, etc.)
  • core/ — runtime, event bus, orchestrator
  • cognition/ — perception, event detection, fusion
  • memory/ — persistence (event logs, context snapshots)
  • agents/ — behavior implementations (summarizers, alerts)
  • workflows/ — higher-level orchestrations
  • schemas/ — shared data models (Signal, Event, Context)
  • config/ — runtime configuration and defaults
• tests/ — automated unit tests

AgentPi supports configuration via:

• YAML file (passed via --config)
• Environment variables using the AGENTPI_ prefix
• CLI override (e.g., --duration)

Example config file (agentpi.yml):

adapter: mock
poll_interval_seconds: 0.5
log_level: INFO
run_duration_seconds: 20

Run with config:

python -m agentpi --config agentpi.yml

Environment variables override defaults:

export AGENTPI_POLL_INTERVAL_SECONDS=0.2
python -m agentpi

python -m pytest

Feel free to add new adapters, agents, workflows, and cognition rules. Keep modules small and preferring async patterns.

This project is released under the MIT License.