Never stop coding. Free MIT AI gateway: one endpoint, 375 providers (150+ free), 1,400+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 900+ contributors
TypeScript a2aai-agentsai-gatewayanthropicclaudeclaude-codeclinecodexcopilotcursordeepseekfree-aigeminikimillm-gatewaymcpopenaiopenai-proxyqwentoken-saver
import test from "node:test";
import assert from "node:assert/strict";
import fs from "node:fs";
import os from "node:os";
import path from "node:path";
const TEST_DATA_DIR = fs.mkdtempSync(
path.join(os.tmpdir(), "omniroute-compaction-input-cap-alignment-")
);
process.env.DATA_DIR = TEST_DATA_DIR;
const core = await import("../../src/lib/db/core.ts");
const { setModelCapabilityOverride } = await import("../../src/lib/db/modelCapabilityOverrides.ts");
const { updateCompressionSettings } = await import("../../src/lib/db/compression.ts");
const { handleChatCore } = await import("../../open-sse/handlers/chatCore.ts");
const TEST_PROVIDER = "test-compaction-prov";
const TEST_MODEL = "test-compaction-model";
const ENV_CONTEXT_LIMIT = "CONTEXT_LENGTH_TEST_COMPACTION_PROV";
const originalFetch = globalThis.fetch;
const originalEnvContext = process.env[ENV_CONTEXT_LIMIT];
async function resetStorage() {
core.resetDbInstance();
fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
}
test.before(async () => {
await resetStorage();
// Window is 8000 tokens, but input-only cap is 4000 tokens
process.env[ENV_CONTEXT_LIMIT] = "8000";
await updateCompressionSettings({
enabled: true,
defaultMode: "off",
});
setModelCapabilityOverride(`${TEST_PROVIDER}/${TEST_MODEL}`, "max_input_tokens", 4000);
globalThis.fetch = async () =>
new Response(
JSON.stringify({
id: "chatcmpl-test",
object: "chat.completion",
created: Date.now(),
model: TEST_MODEL,
choices: [
{
index: 0,
message: { role: "assistant", content: "ok from upstream" },
finish_reason: "stop",
},
],
usage: { prompt_tokens: 3000, completion_tokens: 5, total_tokens: 3005 },
}),
{
status: 200,
headers: { "content-type": "application/json" },
}
);
});
test.after(async () => {
globalThis.fetch = originalFetch;
if (originalEnvContext === undefined) {
delete process.env[ENV_CONTEXT_LIMIT];
} else {
process.env[ENV_CONTEXT_LIMIT] = originalEnvContext;
}
await new Promise((r) => setTimeout(r, 200));
core.resetDbInstance();
fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
});
test("aligns reactive compaction ceiling to max_input_tokens so dead band [cap, window) auto-compacts instead of 400", async () => {
// Generate multi-turn history with ~5,500 estimated tokens (chars/4).
// Fits in window (8000) but exceeds input cap (4000).
// Without the fix, last-resort compaction skipped (5500 < 8000) and enforceOutputTokenBudget rejects with 400.
// With the fix, last-resort targets 4000, prunes older history, fits < 4000 and succeeds.
const turns = [];
for (let i = 0; i < 10; i++) {
turns.push({
role: "user",
content: `Turn ${i} user request: ` + "word ".repeat(250),
});
turns.push({
role: "assistant",
content: `Turn ${i} assistant reply: ` + "answer ".repeat(250),
});
}
const result = await handleChatCore({
body: {
model: TEST_MODEL,
messages: turns,
stream: false,
},
modelInfo: {
provider: TEST_PROVIDER,
model: TEST_MODEL,
extendedContext: false,
},
credentials: {
apiKey: "sk-test",
providerSpecificData: {
baseUrl: "https://test-compaction.example.test",
},
},
clientRawRequest: {
endpoint: "/v1/chat/completions",
body: {
model: TEST_MODEL,
messages: turns,
stream: false,
},
headers: new Headers({ accept: "application/json" }),
},
userAgent: "unit-test",
isCombo: false,
log: {
debug() {},
info() {},
warn() {},
error() {},
},
});
assert.equal(
result.success,
true,
`expected request to succeed via last-resort compaction to input cap, got error: ${
(result as { error?: string }).error
}`
);
});