fix: add missing 'json' keyword to graph memory prompts (fixes #4248) (#4249)

This commit is contained in:
Giulio Leone
2026-03-13 01:38:24 -07:00
committed by GitHub
parent 861cbb7289
commit e95090e116
4 changed files with 669 additions and 1 deletions
+2
View File
@@ -88,6 +88,8 @@ Memory Format:
source -- relationship -- destination
Provide a list of deletion instructions, each specifying the relationship to be deleted.
Respond in JSON format.
`;
export function getDeleteMessages(
+1 -1
View File
@@ -212,7 +212,7 @@ export class MemoryGraph {
[
{
role: "system",
content: `You are a smart assistant who understands entities and their types in a given text. If user message contains self reference such as 'I', 'me', 'my' etc. then use ${filters["userId"]} as the source entity. Extract all the entities from the text. ***DO NOT*** answer the question itself if the given text is a question.`,
content: `You are a smart assistant who understands entities and their types in a given text. If user message contains self reference such as 'I', 'me', 'my' etc. then use ${filters["userId"]} as the source entity. Extract all the entities from the text. ***DO NOT*** answer the question itself if the given text is a question. Respond in JSON format.`,
},
{ role: "user", content: data },
],
@@ -0,0 +1,498 @@
/**
* Regression tests for graph_memory.ts response parsing (issue #4248).
*
* Exercises the three json_object call sites in MemoryGraph with a mocked LLM:
* 1. _retrieveNodesFromData → entity extraction
* 2. _establishNodesRelationsFromData → relation extraction
* 3. _getDeleteEntitiesFromSearchOutput → deletion identification
*
* Covers: malformed LLM responses, missing fields, bad JSON in toolCalls,
* string-only responses, empty tool calls, and prompt construction.
*
* See: https://github.com/mem0ai/mem0/issues/4248
*/
import { MemoryGraph } from "../src/memory/graph_memory";
import { EXTRACT_RELATIONS_PROMPT, getDeleteMessages } from "../src/graphs/utils";
// ---------------------------------------------------------------------------
// Mocks – we replace heavy dependencies so tests run without Neo4j / OpenAI
// ---------------------------------------------------------------------------
// Mock neo4j-driver: provides a fake Driver with a no-op session
jest.mock("neo4j-driver", () => ({
__esModule: true,
default: {
driver: jest.fn(() => ({
session: () => ({
run: jest.fn().mockResolvedValue({ records: [] }),
close: jest.fn(),
}),
})),
auth: { basic: jest.fn() },
},
}));
// Mock factory so constructor doesn't try to instantiate real LLMs / embedders
const mockGenerateResponse = jest.fn();
const mockGenerateChat = jest.fn();
const mockEmbed = jest.fn().mockResolvedValue([0.1, 0.2, 0.3]);
jest.mock("../src/utils/factory", () => ({
LLMFactory: {
create: jest.fn(() => ({
generateResponse: mockGenerateResponse,
generateChat: mockGenerateChat,
})),
},
EmbedderFactory: {
create: jest.fn(() => ({
embed: mockEmbed,
})),
},
}));
// Minimal config that satisfies the MemoryGraph constructor
function makeConfig(overrides: Record<string, any> = {}) {
return {
graphStore: {
config: {
url: "bolt://localhost:7687",
username: "neo4j",
password: "test",
},
...overrides,
},
embedder: { provider: "openai", config: {} },
llm: { provider: "openai", config: {} },
} as any;
}
// Helper to access private methods via `any` cast
function graph(overrides: Record<string, any> = {}): any {
return new MemoryGraph(makeConfig(overrides));
}
const FILTERS = { userId: "test-user" };
beforeEach(() => {
jest.clearAllMocks();
});
// ═══════════════════════════════════════════════════════════════════════════
// 1. _retrieveNodesFromData – entity extraction
// ═══════════════════════════════════════════════════════════════════════════
describe("_retrieveNodesFromData", () => {
it("parses a well-formed extract_entities tool call", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "extract_entities",
arguments: JSON.stringify({
entities: [
{ entity: "Alice", entity_type: "person" },
{ entity: "Pizza", entity_type: "food" },
],
}),
},
],
});
const mg = graph();
const result = await mg._retrieveNodesFromData("Alice likes pizza", FILTERS);
expect(result).toEqual({ alice: "person", pizza: "food" });
});
it("returns empty map when LLM returns a plain string", async () => {
mockGenerateResponse.mockResolvedValueOnce("I am a string, not an object");
const mg = graph();
const result = await mg._retrieveNodesFromData("anything", FILTERS);
expect(result).toEqual({});
});
it("returns empty map when toolCalls is undefined", async () => {
mockGenerateResponse.mockResolvedValueOnce({});
const mg = graph();
const result = await mg._retrieveNodesFromData("anything", FILTERS);
expect(result).toEqual({});
});
it("returns empty map when toolCalls is an empty array", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
const result = await mg._retrieveNodesFromData("anything", FILTERS);
expect(result).toEqual({});
});
it("handles malformed JSON in tool call arguments gracefully", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{ name: "extract_entities", arguments: "NOT VALID JSON {{{" },
],
});
const mg = graph();
// Should not throw — the catch block in the source logs the error
const result = await mg._retrieveNodesFromData("anything", FILTERS);
expect(result).toEqual({});
});
it("handles missing entities array in arguments", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "extract_entities",
arguments: JSON.stringify({ wrong_key: [] }),
},
],
});
const mg = graph();
// args.entities is undefined → for..of on undefined throws → caught
const result = await mg._retrieveNodesFromData("anything", FILTERS);
expect(result).toEqual({});
});
it("skips tool calls with unrelated names", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "some_other_tool",
arguments: JSON.stringify({ entities: [{ entity: "X", entity_type: "Y" }] }),
},
],
});
const mg = graph();
const result = await mg._retrieveNodesFromData("anything", FILTERS);
expect(result).toEqual({});
});
it("normalises entity names to lowercase with underscores", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "extract_entities",
arguments: JSON.stringify({
entities: [{ entity: "New York City", entity_type: "City Name" }],
}),
},
],
});
const mg = graph();
const result = await mg._retrieveNodesFromData("anything", FILTERS);
expect(result).toEqual({ new_york_city: "city_name" });
});
it("passes json_object response format and the correct system prompt", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
await mg._retrieveNodesFromData("test data", FILTERS);
const [messages, responseFormat] = mockGenerateResponse.mock.calls[0];
expect(responseFormat).toEqual({ type: "json_object" });
const systemMsg = messages[0].content as string;
expect(systemMsg.toLowerCase()).toContain("json");
expect(systemMsg).toContain("test-user");
});
});
// ═══════════════════════════════════════════════════════════════════════════
// 2. _establishNodesRelationsFromData – relation extraction
// ═══════════════════════════════════════════════════════════════════════════
describe("_establishNodesRelationsFromData", () => {
it("parses a well-formed establish_relationships tool call", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "establish_relationships",
arguments: JSON.stringify({
entities: [
{ source: "Alice", relationship: "likes", destination: "Pizza" },
],
}),
},
],
});
const mg = graph();
const result = await mg._establishNodesRelationsFromData(
"Alice likes pizza",
FILTERS,
{ alice: "person", pizza: "food" },
);
expect(result).toEqual([
{ source: "alice", relationship: "likes", destination: "pizza" },
]);
});
it("returns empty array when LLM returns a string", async () => {
mockGenerateResponse.mockResolvedValueOnce("just a string");
const mg = graph();
const result = await mg._establishNodesRelationsFromData("x", FILTERS, {});
expect(result).toEqual([]);
});
it("returns empty array when toolCalls is empty", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
const result = await mg._establishNodesRelationsFromData("x", FILTERS, {});
expect(result).toEqual([]);
});
it("returns empty array when entities key is missing from arguments", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "establish_relationships",
arguments: JSON.stringify({ not_entities: [] }),
},
],
});
const mg = graph();
const result = await mg._establishNodesRelationsFromData("x", FILTERS, {});
// args.entities is undefined → falls back to []
expect(result).toEqual([]);
});
it("throws on malformed JSON in tool call arguments (no try/catch in source)", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{ name: "establish_relationships", arguments: "<<BROKEN>>" },
],
});
const mg = graph();
// _establishNodesRelationsFromData does JSON.parse without try/catch
await expect(
mg._establishNodesRelationsFromData("x", FILTERS, {}),
).rejects.toThrow();
});
it("appends JSON format suffix to system prompt (no custom prompt)", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
await mg._establishNodesRelationsFromData("data", FILTERS, { a: "b" });
const [messages, responseFormat] = mockGenerateResponse.mock.calls[0];
expect(responseFormat).toEqual({ type: "json_object" });
const systemContent = messages[0].content as string;
expect(systemContent.toLowerCase()).toContain("json");
expect(systemContent).toContain("test-user");
expect(systemContent).not.toContain("USER_ID");
// CUSTOM_PROMPT placeholder stays when no custom prompt is configured
// (only replaced when config.graphStore.customPrompt is set)
});
it("appends JSON format suffix and custom prompt when configured", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph({ customPrompt: "Focus on food relationships only." });
await mg._establishNodesRelationsFromData("data", FILTERS, {});
const [messages] = mockGenerateResponse.mock.calls[0];
const systemContent = messages[0].content as string;
expect(systemContent.toLowerCase()).toContain("json");
expect(systemContent).toContain("Focus on food relationships only.");
});
});
// ═══════════════════════════════════════════════════════════════════════════
// 3. _getDeleteEntitiesFromSearchOutput – deletion identification
// ═══════════════════════════════════════════════════════════════════════════
describe("_getDeleteEntitiesFromSearchOutput", () => {
const SEARCH_OUTPUT = [
{
source: "alice",
source_id: "1",
relationship: "likes",
relation_id: "r1",
destination: "pizza",
destination_id: "2",
similarity: 0.95,
},
];
it("parses a well-formed delete_graph_memory tool call", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "delete_graph_memory",
arguments: JSON.stringify({
source: "Alice",
relationship: "likes",
destination: "Pizza",
}),
},
],
});
const mg = graph();
const result = await mg._getDeleteEntitiesFromSearchOutput(SEARCH_OUTPUT, "Alice hates pizza", FILTERS);
expect(result).toEqual([
{ source: "alice", relationship: "likes", destination: "pizza" },
]);
});
it("returns empty array when LLM returns a string", async () => {
mockGenerateResponse.mockResolvedValueOnce("string response");
const mg = graph();
const result = await mg._getDeleteEntitiesFromSearchOutput(SEARCH_OUTPUT, "x", FILTERS);
expect(result).toEqual([]);
});
it("returns empty array when no tool calls are present", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
const result = await mg._getDeleteEntitiesFromSearchOutput(SEARCH_OUTPUT, "x", FILTERS);
expect(result).toEqual([]);
});
it("skips non-delete_graph_memory tool calls", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "noop",
arguments: JSON.stringify({}),
},
],
});
const mg = graph();
const result = await mg._getDeleteEntitiesFromSearchOutput(SEARCH_OUTPUT, "x", FILTERS);
expect(result).toEqual([]);
});
it("collects multiple delete tool calls", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "delete_graph_memory",
arguments: JSON.stringify({ source: "A", relationship: "r1", destination: "B" }),
},
{
name: "delete_graph_memory",
arguments: JSON.stringify({ source: "C", relationship: "r2", destination: "D" }),
},
],
});
const mg = graph();
const result = await mg._getDeleteEntitiesFromSearchOutput(SEARCH_OUTPUT, "x", FILTERS);
expect(result).toHaveLength(2);
expect(result[0].source).toBe("a");
expect(result[1].source).toBe("c");
});
it("passes json_object format and includes 'json' in system prompt", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
await mg._getDeleteEntitiesFromSearchOutput(SEARCH_OUTPUT, "data", FILTERS);
const [messages, responseFormat] = mockGenerateResponse.mock.calls[0];
expect(responseFormat).toEqual({ type: "json_object" });
const systemContent = messages[0].content as string;
expect(systemContent.toLowerCase()).toContain("json");
expect(systemContent).toContain("test-user");
expect(systemContent).not.toContain("USER_ID");
});
it("handles empty searchOutput array", async () => {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
const result = await mg._getDeleteEntitiesFromSearchOutput([], "data", FILTERS);
expect(result).toEqual([]);
});
});
// ═══════════════════════════════════════════════════════════════════════════
// 4. Prompt construction — JSON keyword present in every json_object site
// ═══════════════════════════════════════════════════════════════════════════
describe("Prompt construction — all json_object sites include 'json'", () => {
it("_retrieveNodesFromData system message includes 'json' for any userId", async () => {
for (const userId of ["", "user-1", "special<>chars", "ユーザー"]) {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
await mg._retrieveNodesFromData("test", { userId });
const systemMsg = mockGenerateResponse.mock.calls.at(-1)![0][0].content;
expect(systemMsg.toLowerCase()).toContain("json");
}
});
it("_establishNodesRelationsFromData system message includes 'json' for any userId", async () => {
for (const userId of ["", "user-1", "special<>chars"]) {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
await mg._establishNodesRelationsFromData("test", { userId }, {});
const systemMsg = mockGenerateResponse.mock.calls.at(-1)![0][0].content;
expect(systemMsg.toLowerCase()).toContain("json");
}
});
it("_getDeleteEntitiesFromSearchOutput system message includes 'json' for any userId", async () => {
for (const userId of ["", "user-1", "special<>chars"]) {
mockGenerateResponse.mockResolvedValueOnce({ toolCalls: [] });
const mg = graph();
await mg._getDeleteEntitiesFromSearchOutput([], "test", { userId });
const systemMsg = mockGenerateResponse.mock.calls.at(-1)![0][0].content;
expect(systemMsg.toLowerCase()).toContain("json");
}
});
});
// ═══════════════════════════════════════════════════════════════════════════
// 5. Edge cases – malformed entity fields in _removeSpacesFromEntities
// ═══════════════════════════════════════════════════════════════════════════
describe("_removeSpacesFromEntities (via _establishNodesRelationsFromData)", () => {
it("normalises spaces and case in entity source/relationship/destination", async () => {
mockGenerateResponse.mockResolvedValueOnce({
toolCalls: [
{
name: "establish_relationships",
arguments: JSON.stringify({
entities: [
{ source: "New York", relationship: "Capital Of", destination: "United States" },
],
}),
},
],
});
const mg = graph();
const result = await mg._establishNodesRelationsFromData("test", FILTERS, {});
expect(result).toEqual([
{ source: "new_york", relationship: "capital_of", destination: "united_states" },
]);
});
});
+168
View File
@@ -0,0 +1,168 @@
import {
DELETE_RELATIONS_SYSTEM_PROMPT,
EXTRACT_RELATIONS_PROMPT,
UPDATE_GRAPH_PROMPT,
getDeleteMessages,
formatEntities,
} from "../src/graphs/utils";
/**
* Regression tests for graph prompts (issue #4248).
*
* When response_format: { type: "json_object" } is used, OpenAI requires
* the word "json" (case-insensitive) to appear in at least one message.
* Missing it produces a 400 error.
*
* Three call sites use json_object today:
* 1. _getDeleteEntitiesFromSearchOutput → DELETE_RELATIONS_SYSTEM_PROMPT
* 2. _retrieveNodesFromData → inline prompt (graph_memory.ts)
* 3. _getRelatedEntities → EXTRACT_RELATIONS_PROMPT + suffix
*
* See: https://github.com/mem0ai/mem0/issues/4248
*/
// ─── JSON keyword presence ────────────────────────────────────────────────────
describe("Graph prompts — JSON keyword requirement", () => {
it("DELETE_RELATIONS_SYSTEM_PROMPT contains 'json'", () => {
expect(DELETE_RELATIONS_SYSTEM_PROMPT.toLowerCase()).toContain("json");
});
it("EXTRACT_RELATIONS_PROMPT produces a message containing 'json' once the suffix is appended", () => {
// graph_memory.ts appends "\nPlease provide your response in JSON format."
const withSuffix =
EXTRACT_RELATIONS_PROMPT + "\nPlease provide your response in JSON format.";
expect(withSuffix.toLowerCase()).toContain("json");
});
it("getDeleteMessages system message contains 'json' after USER_ID substitution", () => {
const [systemContent] = getDeleteMessages(
"alice -- loves -- pizza",
"Alice now hates pizza",
"user-42",
);
expect(systemContent.toLowerCase()).toContain("json");
});
it("entity extraction inline prompt contains 'json' (simulated from graph_memory.ts)", () => {
// Mirrors the template in _retrieveNodesFromData()
const userId = "user-1";
const prompt = `You are a smart assistant who understands entities and their types in a given text. If user message contains self reference such as 'I', 'me', 'my' etc. then use ${userId} as the source entity. Extract all the entities from the text. ***DO NOT*** answer the question itself if the given text is a question. Respond in JSON format.`;
expect(prompt.toLowerCase()).toContain("json");
});
});
// ─── getDeleteMessages ────────────────────────────────────────────────────────
describe("getDeleteMessages", () => {
it("replaces USER_ID with the provided userId in the system prompt", () => {
const [system] = getDeleteMessages("mem", "data", "alice-123");
expect(system).toContain("alice-123");
expect(system).not.toContain("USER_ID");
});
it("includes existing memories and new data in the user prompt", () => {
const existing = "bob -- knows -- carol";
const newData = "Bob no longer knows Carol";
const [, user] = getDeleteMessages(existing, newData, "u1");
expect(user).toContain(existing);
expect(user).toContain(newData);
});
it("returns a 2-tuple [system, user]", () => {
const result = getDeleteMessages("a", "b", "c");
expect(result).toHaveLength(2);
expect(typeof result[0]).toBe("string");
expect(typeof result[1]).toBe("string");
});
// — Malformed / edge-case inputs —
it("handles empty strings without throwing", () => {
expect(() => getDeleteMessages("", "", "")).not.toThrow();
const [system, user] = getDeleteMessages("", "", "");
expect(system.toLowerCase()).toContain("json");
expect(typeof user).toBe("string");
});
it("handles special characters in userId (e.g. angle brackets, quotes)", () => {
const [system] = getDeleteMessages("mem", "data", '<script>alert("xss")</script>');
expect(system).toContain('<script>alert("xss")</script>');
expect(system).not.toContain("USER_ID");
});
it("handles unicode input", () => {
const [system, user] = getDeleteMessages("日本語メモリ", "新しい情報", "ユーザー1");
expect(system).toContain("ユーザー1");
expect(user).toContain("日本語メモリ");
expect(user).toContain("新しい情報");
});
it("handles very long input strings", () => {
const longStr = "x".repeat(100_000);
expect(() => getDeleteMessages(longStr, longStr, "u")).not.toThrow();
const [system] = getDeleteMessages(longStr, longStr, "u");
expect(system.toLowerCase()).toContain("json");
});
});
// ─── formatEntities ───────────────────────────────────────────────────────────
describe("formatEntities", () => {
it("formats a single entity triplet", () => {
const result = formatEntities([
{ source: "Alice", relationship: "knows", destination: "Bob" },
]);
expect(result).toBe("Alice -- knows -- Bob");
});
it("joins multiple entities with newlines", () => {
const result = formatEntities([
{ source: "A", relationship: "r1", destination: "B" },
{ source: "C", relationship: "r2", destination: "D" },
]);
expect(result).toBe("A -- r1 -- B\nC -- r2 -- D");
});
it("returns empty string for empty array", () => {
expect(formatEntities([])).toBe("");
});
it("preserves special characters in entity fields", () => {
const result = formatEntities([
{ source: "O'Brien", relationship: "said \"hello\"", destination: "café" },
]);
expect(result).toContain("O'Brien");
expect(result).toContain('said "hello"');
expect(result).toContain("café");
});
});
// ─── Prompt structural invariants ─────────────────────────────────────────────
describe("Prompt structural invariants", () => {
it("DELETE_RELATIONS_SYSTEM_PROMPT contains USER_ID placeholder", () => {
expect(DELETE_RELATIONS_SYSTEM_PROMPT).toContain("USER_ID");
});
it("EXTRACT_RELATIONS_PROMPT contains USER_ID placeholder", () => {
expect(EXTRACT_RELATIONS_PROMPT).toContain("USER_ID");
});
it("EXTRACT_RELATIONS_PROMPT contains CUSTOM_PROMPT placeholder", () => {
expect(EXTRACT_RELATIONS_PROMPT).toContain("CUSTOM_PROMPT");
});
it("UPDATE_GRAPH_PROMPT contains memory template placeholders", () => {
expect(UPDATE_GRAPH_PROMPT).toContain("{existing_memories}");
expect(UPDATE_GRAPH_PROMPT).toContain("{new_memories}");
});
it("DELETE_RELATIONS_SYSTEM_PROMPT is non-empty and reasonably sized", () => {
expect(DELETE_RELATIONS_SYSTEM_PROMPT.length).toBeGreaterThan(100);
});
it("EXTRACT_RELATIONS_PROMPT is non-empty and reasonably sized", () => {
expect(EXTRACT_RELATIONS_PROMPT.length).toBeGreaterThan(100);
});
});