Sagewai documentation

Sagewai is a Python SDK for building agents. Install it with pip install sagewai, write an agent in a few lines, and run it. When you need more — a web admin interface, a pool of remote workers, full observability, or hardened security — the platform components are there to adopt incrementally.

The mental model

SDK first. sagewai is a Python library. You import it, configure an agent, and call it. No server required, no account required, no API key required for local models. Everything starts here.

Platform second. When you outgrow a single process you add the pieces you need:

  • Admin server — a FastAPI server (sagewai[fastapi]) that exposes a web UI and REST API for managing agents, providers, and runs.
  • Fleet workers — remote processes that register with the server and execute tasks; scale horizontally, project-isolated.
  • Observatory — OpenTelemetry-based observability: metrics, logs, and traces routed through VictoriaMetrics and Grafana.
  • Sealed security — Tier-1/Tier-2 trust boundaries, JIT human-in-the-loop approvals, and vault-backed credential scoping.
  • Autopilot — state a goal in plain English; the platform designs, provisions, and runs the agents that deliver it.
  • Training Loop — collect runs, curate them, export fine-tuning datasets, and deploy your own model via Ollama.

None of these require the others. Add what you need, skip what you do not.

Start here

  1. Installation — pip install sagewai, extras, and the verify step.
  2. Prerequisites — Python 3.10+, Docker (sandbox path only), hardware profiles.
  3. Your first agent — build and run an agent from scratch.

Find your way around

  • Core Concepts — agents, strategies, memory, workflows, directives, safety.
  • Platform — admin server, autopilot, fleet, observatory, training loop, Sealed security.
  • Guides — task-focused how-tos: multi-agent workflows, fine-tuning, deployment, integrations.
  • Tutorials — end-to-end walkthroughs that build complete systems step by step.
  • Architecture — runtime topology, security tiers, execution modes, sandbox backends.
  • Reference — numbered example file catalogue, full Python SDK API reference.