Quickstart

This is the fastest path to a running Thodare. Five minutes from npm install to a workflow you can curl.

Prerequisites

• Node 22+ (fetch, node:sqlite are built-in).
• Postgres reachable locally — the engine’s source of truth.
• A boot of @thodare/api somewhere your terminal can reach.

If you don’t have an API instance running yet, see Reference example for a 30-line TypeScript file that boots one in your repo. Then come back here.

1. Install the CLI

npm install -g @thodare/cli
# or
pnpm add -g @thodare/cli

2. Sign in (or sign up)

thodare login --api http://localhost:3000
# → Email: you@example.com
# → Password: ********
# → Signed up as you@example.com
# → Active org: you-7f2k (org_xxxx)
# → API key: thd_… (saved to ~/.thodare/credentials.json)

The CLI runs the full bootstrap in one shot: sign in or sign up → ensure your account has an organization (auto-creates one if not) → set it active → mint an API key → save credentials.

If the database is empty (fresh deploy), thodare login won’t work — nobody can sign up because there’s no admin route by default. Use the bootstrap admin link instead.

3. Verify

thodare whoami
# you@example.com
# org: you-7f2k (org_xxxx)
# api: http://localhost:3000

curl -H "Authorization: Bearer $(thodare token)" \
     http://localhost:3000/api/connectors?detail=summary | jq

You should see the connector catalog — that’s what your LLM will read from its system prompt.

4. Create a workflow

URL=http://localhost:3000
H="Authorization: Bearer $(thodare token)"

WF=$(curl -sX POST "$URL/api/workflows" -H "$H" -H 'content-type: application/json' -d '{}')
WFID=$(echo "$WF" | jq -r .id)
VER=$(echo "$WF" | jq -r .version)

5. Patch it (intentionally broken — to see how feedback works)

curl -sX POST "$URL/api/workflows/$WFID/operations" \
  -H "$H" -H "If-Match: $VER" -H 'content-type: application/json' \
  -d '{"ops":[
    {"operation_type":"add","block_id":"trg","type":"trigger_webhook","params":{}},
    {"operation_type":"add","block_id":"g","type":"slak","params":{"channel":"#sales","text":"hi"}}
  ]}' | jq '{ok, version, summary, skipped_items}'

The response’s skipped_items[] tells the LLM exactly what’s wrong (typo: slak → slack). This is the load-bearing primitive — see The patch loop.

6. Fix and run

VER=$(curl -s "$URL/api/workflows/$WFID" -H "$H" | jq .version)

curl -sX POST "$URL/api/workflows/$WFID/operations" \
  -H "$H" -H "If-Match: $VER" -H 'content-type: application/json' \
  -d '{"ops":[
    {"operation_type":"add","block_id":"g","type":"slack","params":{"channel":"#sales","text":"hi"}},
    {"operation_type":"connect","block_id":"trg","target_block_id":"g"}
  ]}' >/dev/null

RUN=$(curl -sX POST "$URL/api/workflows/$WFID/run" -H "$H" -H 'content-type: application/json' \
  -d '{"input":{"hello":"world"}}')
RUNID=$(echo "$RUN" | jq -r .runId)

while true; do
  STATE=$(curl -s "$URL/api/runs/$RUNID" -H "$H" | jq -r .state)
  echo "state=$STATE"
  [ "$STATE" = "completed" ] || [ "$STATE" = "failed" ] && break
  sleep 0.5
done

curl -s "$URL/api/runs/$RUNID" -H "$H" | jq .output

Done. You patched, ran, and read the output of a durable workflow.

What just happened

• thodare login collapsed sign-up + org + key into one command.
• The first patch had a typo; skipped_items[] carried the structured rejection — feedable directly to an LLM.
• The fix-up patch landed cleanly because the LLM had structured feedback, not just a 400.
• Dispatch packed the workflow JSON into the run input (so a later patch wouldn’t disturb the run) and the durable runtime took over.

Next

• Build your first workflow — same flow, with TypeScript.
• Define a connector — the building block.
• The patch loop — why skip-don’t-reject works.