| name | setup-pi-engram-extensions |
| description | Full workspace bootstrap for the pi-engram-extensions package. Checks prerequisites (pi, node, engram CLI), verifies the package is installed and extensions are loading, then runs the model-failover setup to configure failover.json and model-routing.json. Use this when installing on a new machine, after a subscription change, or when handing off to a new agent or developer. |
Setup Pi Engram Extensions
Bootstraps the full @vincents-ai/pi-engram-extensions package. Covers:
- Prerequisite checks (pi, node, engram CLI)
- Package install verification
- Model failover configuration
- Engram workspace initialisation
Read extensions-overview.md for a description of each extension.
Step 0 — Install the package (if not already done)
pi install -l git:github.com/vincents-ai/pi-engram-extensions
pi install git:github.com/vincents-ai/pi-engram-extensions
Then restart pi or run /reload.
Step 1 — Check prerequisites
Run the check script:
bash ./scripts/check-prerequisites.sh
This checks:
pi, node, engram are in PATH
.pi/failover.json and .pi/model-routing.json exist and are valid JSON
- Engram workspace is initialised (
.engram/ present)
~/.pi/agent/auth.json OAuth providers and key env vars
If there are failures (✗): fix them before continuing. Warnings (⚠) are non-blocking.
Common fixes:
| Failure | Fix |
|---|
engram not found | /skill:install-engram |
failover.json not found | Continue to Step 3 — it will be created |
.engram/ not found | engram setup workspace in the project root |
pi not found | Install pi from https://github.com/mariozechner/pi-coding-agent |
After running, store the result:
engram_context_create
title="Prerequisites check: <pass/fail count>"
content="<paste summary line from script output>"
relevance="medium"
engram_relationship_create source_id=<context UUID> source_type="context"
target_id=<task UUID> target_type="task"
relationship_type="relates_to" agent="<name>"
Step 2 — Bootstrap the engram workspace (if not done)
If .engram/ was missing, initialise it now:
engram setup workspace
engram setup agent --name "Claude" --agent-type implementation
engram skills setup
engram validate hook install
See /skill:install-engram for the full bootstrap reference.
Step 3 — Configure model failover
Load the setup-model-failover skill and follow its full protocol:
/skill:setup-model-failover
That skill will:
- Detect which AI providers are configured (OAuth tokens + API key env vars)
- Ask you to confirm which subscriptions are active
- Write
.pi/failover.json and .pi/model-routing.json
Do not skip this step — the engram-agents and engram-orchestrator extensions use these configs to pick the right model for every dispatched subagent.
Step 4 — Verify extensions are loading
Run /reload in pi (or restart), then test each extension:
| Check | Command | Expected |
|---|
| engram-tools | engram_ask query="test" | Returns results or "no results" |
| engram-agents | /agent-list | Lists 171 agent personas |
| model-failover | /failover-status | Shows current model + priority list |
| engram-orchestrator | /orchestrate help | Shows orchestration usage |
Step 5 — Summary and handoff
Report to the user:
- Which extensions loaded successfully
- Current active model and failover model count
- Engram workspace status (
engram next, engram validate check)
- Any remaining warnings or next steps
If handing off to a new agent or developer, also include:
- Skill to re-run if subscriptions change:
/skill:setup-model-failover
- Skill to re-run for full setup on a new machine:
/skill:setup-pi-engram-extensions
- Where configs live:
.pi/failover.json, .pi/model-routing.json
- Prior decisions in engram:
engram ask query "model routing failover extensions"