Skip to content

Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.

Refuse to start the built-in Bot without a model key, so the healthcheck stops passing a Bot that cannot answer (#99)

Jeongwan Jeon7642adebef3e
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 2a14feea..14225907 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -66,6 +66,10 @@ jobs:
with:
bun-version: 1.3.14
- run: bun install --frozen-lockfile
+ # The Bots are not root workspaces and each keeps its own lockfile. The startup tests spawn
+ # the real entrypoint, so its dependencies have to be installed as well.
+ - run: bun install --frozen-lockfile
+ working-directory: agent-bot
# Not the db:migrate script: that one loads ../.env, which does not exist in CI. DATABASE_URL
# comes from the job env instead, which drizzle.config.ts already reads.
- run: bunx drizzle-kit migrate --config=drizzle.config.ts
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5375f331..285a1411 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -30,6 +30,10 @@ digit. The same rule container and volume names have always followed. A deployme
`AUDIT_RETENTION_DAYS` is new and unset, which keeps the audit trail forever, as before. Set it to a
whole number of days to have old rows removed.
+The built-in Bot refuses to start without `OPENAI_API_KEY`. It used to start, report healthy, and
+then fail every conversation, so a missing key looked like a working deployment. The LangGraph Bot
+already refused the same way.
+
Sessions survive and nobody signs in again.
### Added
diff --git a/agent-bot/src/index.ts b/agent-bot/src/index.ts
index 3fecd861..e48d318d 100644
--- a/agent-bot/src/index.ts
+++ b/agent-bot/src/index.ts
@@ -45,8 +45,25 @@ const MODEL = process.env.BOT_MODEL ?? "gpt-5.5";
*/
const BASE_URL = process.env.OPENAI_BASE_URL?.trim() || undefined;
+/**
+ * The key that model is answered with, checked at startup rather than on the first conversation.
+ *
+ * Without the check the Bot starts, answers the healthcheck, and then fails every run, so the
+ * compose healthcheck reports a Bot that cannot answer as healthy. The LangGraph Bot already
+ * refuses to start without its provider's key, and this file already refuses without its token
+ * above; the model key was the one configuration that escaped the same posture. A missing key
+ * should fail in front of whoever is deploying, not in front of whoever is asking.
+ */
+const API_KEY = process.env.OPENAI_API_KEY?.trim();
+if (!API_KEY) {
+ console.error(
+ "OPENAI_API_KEY is not set. This Bot cannot answer without a model.",
+ );
+ process.exit(1);
+}
+
const openai = new OpenAI({
- apiKey: process.env.OPENAI_API_KEY,
+ apiKey: API_KEY,
baseURL: BASE_URL,
});
diff --git a/tests/agent-bot.test.ts b/tests/agent-bot.test.ts
new file mode 100644
index 00000000..201c72a1
--- /dev/null
+++ b/tests/agent-bot.test.ts
@@ -0,0 +1,57 @@
+import { expect, test } from "bun:test";
+import { join } from "node:path";
+
+/**
+ * The compose healthcheck asks `/health`, and `/health` answers without consulting the model key.
+ * The refusal to run without one therefore has to happen before the server listens: a Bot that
+ * cannot answer should not be running, let alone reporting healthy. These spawn the real
+ * entrypoint with the environment compose would hand it.
+ */
+
+async function startBot(environment: Record<string, string>) {
+ const proc = Bun.spawn(
+ ["bun", join(import.meta.dir, "..", "agent-bot", "src", "index.ts")],
+ {
+ // Every variable the repository's `.env` could inject is named explicitly: bun loads that
+ // file into the child, and a leaked token or key would let a configuration under test pass
+ // a check it is supposed to fail.
+ env: {
+ PATH: process.env.PATH ?? "",
+ MANAGED_AGENT_TOKEN: "",
+ OPENAI_API_KEY: "",
+ ...environment,
+ },
+ stdout: "pipe",
+ stderr: "pipe",
+ },
+ );
+
+ // Both configurations under test exit before the server listens. If one reaches `serve` anyway,
+ // kill it so the test fails on the missing refusal rather than on bun's test timeout.
+ const killer = setTimeout(() => proc.kill(), 5_000);
+ const exitCode = await proc.exited;
+ clearTimeout(killer);
+ const stderr = await new Response(proc.stderr).text();
+ return { exitCode, stderr };
+}
+
+test("agent-bot refuses to start when the model key is empty", async () => {
+ // Compose passes `${OPENAI_API_KEY:-}`, so an unset key arrives as an empty
+ // string rather than missing altogether.
+ const { exitCode, stderr } = await startBot({
+ MANAGED_AGENT_TOKEN: "test-token",
+ OPENAI_API_KEY: "",
+ });
+
+ expect(exitCode).toBe(1);
+ expect(stderr).toContain("OPENAI_API_KEY is not set");
+});
+
+test("agent-bot still refuses to start without its server token", async () => {
+ const { exitCode, stderr } = await startBot({
+ OPENAI_API_KEY: "sk-test",
+ });
+
+ expect(exitCode).toBe(1);
+ expect(stderr).toContain("MANAGED_AGENT_TOKEN is not set");
+});