tinyhumansai/openhuman · error · Error

core not reachable at ${RPC_URL} — start it with `cargo run

Error message

core not reachable at ${RPC_URL} — start it with `cargo run --bin openhuman -- serve`. (${err.message})

What it means

main() sanity-checks connectivity with a single openhuman.health_snapshot RPC before doing any work; when that first call fails for any reason (ECONNREFUSED, HTTP error like 287, or a JSON-RPC error like 288), the script rethrows wrapped with this hint naming the startup command.

Source

Thrown at scripts/test-memory-email-ingest.mjs:112

    threads.get(tid).push(m);
  }
  return [...threads.entries()].map(([threadId, msgs]) => {
    msgs.sort((a, b) => parseEmailDate(a.date) - parseEmailDate(b.date));
    return {
      threadId,
      subject: msgs[0]?.subject || "(no subject)",
      messages: msgs.map(toEmailMessage),
    };
  });
}

async function main() {
  console.log(`[memory-email-ingest] fixture=${FIXTURE}`);
  console.log(`[memory-email-ingest] rpc_url=${RPC_URL}`);

  // Sanity-check that the core is up.
  await rpc("openhuman.health_snapshot", {}).catch((err) => {
    throw new Error(
      `core not reachable at ${RPC_URL} — start it with \`cargo run --bin openhuman -- serve\`. (${err.message})`,
    );
  });

  const raw = await readFile(FIXTURE, "utf8");
  const inbox = JSON.parse(raw);
  const messages = Array.isArray(inbox.messages) ? inbox.messages : [];
  if (messages.length === 0) {
    console.error("[memory-email-ingest] no messages in fixture, nothing to do");
    process.exit(1);
  }
  console.log(`[memory-email-ingest] loaded ${messages.length} email(s)`);

  const threads = groupByThread(messages);
  console.log(`[memory-email-ingest] grouped into ${threads.length} thread(s)`);

  let chunksWritten = 0;
  let chunksDropped = 0;

View on GitHub (pinned to a221052e0d)

Solutions

Start the core: cargo run --bin openhuman -- serve (release tree: ./target/debug/openhuman-core serve) and wait for its listening log
Verify liveness on the same base: curl http://127.0.0.1:7810/health
Align RPC_URL with the port the core actually logged; the token for a standalone core lives in {workspace}/core.token

Example fix

# before
node scripts/test-memory-email-ingest.mjs   # core not reachable at http://127.0.0.1:7810/rpc

# after
cargo run --bin openhuman -- serve &        # wait for the listening line
RPC_URL='http://127.0.0.1:7810/rpc' node scripts/test-memory-email-ingest.mjs

Defensive patterns

Strategy: validation

Validate before calling

async function waitForCore(url, tries = 30) {
  for (let i = 0; i < tries; i++) {
    try { const r = await fetch(url.replace(/\/rpc$/, '/health')); if (r.ok) return; } catch {}
    await new Promise((r) => setTimeout(r, 1000));
  }
  throw new Error(`core never became healthy at ${url}`);
}

Prevention

start the core before the ingest script; gate on /health
keep RPC_URL and token in one sourced env file
treat any health_snapshot failure as environment, not fixture, trouble first

When it happens

Trigger: No core listening on RPC_URL (default http://127.0.0.1:7810/rpc); core on a different port; core up but health_snapshot itself erroring (JSON-RPC failure still lands here via err.message).

Common situations: Forgot to run `cargo run --bin openhuman -- serve` in a separate terminal; core crashed between two script runs; stale RPC_URL exported from an earlier session pointing at a dead port.

Related errors

AI-assisted analysis of tinyhumansai/openhuman@a221052e0d (2026-08-16). Data as JSON: /api/errors/2c8b537d34086d81. Report an issue: GitHub.