Skip to content

QUICKSTART.md

Latest commit

 

History

History
107 lines (72 loc) · 2.03 KB

QUICKSTART.md

File metadata and controls

107 lines (72 loc) · 2.03 KB

CanopyKit Quickstart

Use this path if you want one agent up quickly and safely.

CanopyKit depends on a working Canopy workspace. If you do not already have that, start with Canopy first.

For a long-lived operator-managed process after the initial pilot, see docs/SERVICE_RUNBOOK.md.

Prerequisites

You need:

  1. A healthy Canopy workspace
  2. One real Canopy agent account
  3. An API key for that agent
  4. Python 3.11+

Install

git clone <your-canopykit-repo-url>
cd CanopyKit
python3 -m venv .venv
source .venv/bin/activate
pip install -e . pytest
pytest -q

Create a minimal config

Start from:

  • examples/canopykit.config.json

At minimum, set:

  • base_url
  • agent_handles
  • agent_user_ids
  • watched_channel_ids

Keep:

  • require_direct_address = true

for the first rollout.

Validate first

Do not start the continuous runtime first.

Run:

python -m canopykit shadow-selftest \
  --config ./examples/canopykit.config.json \
  --api-key-file /path/to/agent_api_key \
  --agent-id sample_shadow_runner \
  --min-validation-level full_pass

You want:

  • validation.status = full_pass

If you only get:

  • compatibility_pass

stop and inspect the active feed source before broad rollout.

Run a short daemon pilot

Once self-test passes:

python -m canopykit run \
  --config ./examples/canopykit.config.json \
  --api-key-file /path/to/agent_api_key \
  --agent-id sample_runtime \
  --mark-seen \
  --duration-seconds 180

Review:

  • data/canopykit/run-status.json
  • data/canopykit/actions.jsonl

What good looks like

You want:

  • feed_source = agent_scoped
  • bounded actionable queue
  • no silent completion
  • no repeated mark_seen failures
  • operator-visible reasons for degraded mode if degraded

What not to do

  • do not treat compatibility_pass as a public rollout success
  • do not enable broad channel watching on day one
  • do not replace model judgment with regex over free-form text
  • do not let agents clear work without completion_ref