Platform · autopilot

Autopilot — describe the goal, we design the agents.

Mission, blueprint, hosted service. Declare the outcome; Autopilot decomposes the goal into agents, picks tools, and captures the trace (linear plans today; branched plans in progress). The Curator turns successful runs into JSONL — the seed corpus for a future fine-tune. Goals in, agents out, training data you can build on.

from sagewai.autopilot import Mission

mission = Mission("triage and respond to support emails")
result = mission.run()  # the run trace is captured for the Curator

→ Example 36 — Autopilot training loop Run the 5-minute quickstart

The rest of the platform

SDK — the harness that works with any LLM.
About $0.07 per 100 support emails triaged on Haiku 4.5
Fleet — workers, dispatch, scoped routing.
24 workers / 100 tasks / 0 cross-tenant leaks (CI-gated)
Observatory — show your CFO where the AI money goes.
Real OTel + Grafana, populated by mixed-tenant load
Training Loop — from juggernauts to your own model.
Capture runs, fine-tune a small model, deploy it locally via Ollama
Sealed — every customer secret, scoped and scrubbed.
0 customer secrets reach the worker host (CI-gated)