Skip to content

B3: pre-allocate assistant row at llm_call_started boundary #32326

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
dvargasfuertes merged 4 commits into from
May 29, 2026
Merged

B3: pre-allocate assistant row at llm_call_started boundary #32326

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
5749e25
B3: pre-allocate assistant row at llm_call_started boundary
vellum-apollo-bot[bot] May 27, 2026
8d9c6fe
B3 review iteration: restore indexing, cleanup paths, llm_call_starte…
vellum-apollo-bot[bot] May 28, 2026
3e03247
Merge remote-tracking branch 'origin/main' into apollo/agent-loop-pre…
vellum-apollo-bot[bot] May 28, 2026
cb7733e
test(agent-loop-exit-reason): include llm_call_started in max_tokens …
vellum-apollo-bot[bot] May 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions assistant/src/__tests__/agent-loop-exit-reason.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -167,6 +167,7 @@ describe("AgentLoop exit-reason instrumentation", () => {
});

expect(events.map((e) => e.type)).toEqual([
"llm_call_started",
"usage",
"max_tokens_reached",
"message_complete",
Expand Down
4 changes: 4 additions & 0 deletions assistant/src/__tests__/conversation-abort-tool-results.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -139,6 +139,7 @@ mock.module("../memory/conversation-crud.js", () => ({
getMessageById: () => null,
getLastUserTimestampBefore: () => 0,
reserveMessage: mock(async () => ({ id: "msg-reserve" })),
updateMessageContent: mock(() => {}),
}));

mock.module("../memory/conversation-queries.js", () => ({
Expand Down Expand Up @@ -193,6 +194,9 @@ mock.module("../agent/loop.js", () => ({
onEvent: (event: AgentEvent) => void,
_signal?: AbortSignal,
): Promise<Message[]> {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
const history = [...messages];

// Simulate provider response with 2 tool_use blocks
Expand Down
33 changes: 33 additions & 0 deletions assistant/src/__tests__/conversation-agent-loop-overflow.test.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -736,6 +736,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {

let agentLoopCallCount = 0;
const agentLoopRun: AgentLoopRun = async (messages, onEvent) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
agentLoopCallCount++;
if (agentLoopCallCount === 1) {
// Simulate: agent makes progress (tool calls + results added)
Expand Down Expand Up @@ -915,6 +918,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
};

const agentLoopRun: AgentLoopRun = async (messages, onEvent) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
callCount++;
if (callCount === 1) {
// Provider rejects with "prompt is too long: 242201 tokens > 200000"
Expand Down Expand Up @@ -1037,6 +1043,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
};

const agentLoopRun: AgentLoopRun = async (messages, onEvent) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
callCount++;
if (callCount === 1) {
// Provider rejects: actual tokens 242201, way above estimate of 185k
Expand Down Expand Up @@ -1162,6 +1171,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
};

const agentLoopRun: AgentLoopRun = async (messages, onEvent) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
callCount++;
onEvent({
type: "message_complete",
Expand Down Expand Up @@ -1250,6 +1262,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {

let agentLoopCallCount = 0;
const agentLoopRun: AgentLoopRun = async (messages, onEvent) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
agentLoopCallCount++;
if (agentLoopCallCount === 1) {
// Agent makes progress (tool calls succeed, messages grow)
Expand Down Expand Up @@ -1441,6 +1456,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
_requestId,
onCheckpoint,
) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
agentLoopCallCount++;

if (agentLoopCallCount === 1) {
Expand Down Expand Up @@ -1626,6 +1644,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
_requestId,
onCheckpoint,
) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
agentLoopCallCount++;

if (agentLoopCallCount === 1) {
Expand Down Expand Up @@ -1801,6 +1822,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
_requestId,
onCheckpoint,
) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
agentLoopCallCount++;

// Every call: simulate tool progress then yield at checkpoint
Expand Down Expand Up @@ -1962,6 +1986,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
_requestId,
onCheckpoint,
) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
agentLoopCallCount++;

const withProgress: Message[] = [
Expand Down Expand Up @@ -2194,6 +2221,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
};

const agentLoopRun: AgentLoopRun = async (messages, onEvent) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
onEvent({
type: "message_complete",
message: {
Expand Down Expand Up @@ -2297,6 +2327,9 @@ describe("session-agent-loop overflow recovery (JARVIS-110)", () => {
_requestId,
onCheckpoint,
) => {
// Prime the assistant row anchor — production code emits this from
// `AgentLoop.run` just before `provider.sendMessage`.
await onEvent({ type: "llm_call_started" });
agentLoopCallCount++;

const withProgress: Message[] = [
Expand Down
Loading
Loading