Skip to content

fix(client): record conversation id with canonical cross-format history #817

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

Closed
deepujain wants to merge 2 commits into from
Closed

fix(client): record conversation id with canonical cross-format history #817

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5d93015
fix(client): record conversation id with canonical cross-format history
deepujain Sep 22, 2026
676f55c
fix(client): refresh materialized conversation history on repeat store
deepujain Sep 22, 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
217 changes: 209 additions & 8 deletions crates/libsy-llm-client/src/run.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -518,7 +518,14 @@ impl StateOwners {
}
}
if let Some(id) = conversation_id {
self.by_id.entry(id.to_owned()).or_insert(state);
// Materialized canonical history always describes the conversation's
// latest state, so a repeat store replaces the entry; provider-owned
// state keeps first-writer routing pins.
if materialized {
self.by_id.insert(id.to_owned(), state);
} else {
self.by_id.entry(id.to_owned()).or_insert(state);
}
}
Ok(())
}
Expand Down Expand Up @@ -719,8 +726,14 @@ impl ClientRouter {
.map_err(|error| LibsyError::client_call(model.clone(), error))?;
}
} else if let Some(input) = &canonical_input {
self.remember_canonical_response(&agg, &model, store, input)
.map_err(|error| LibsyError::client_call(model.clone(), error))?;
self.remember_canonical_response(
&agg,
&model,
store,
input,
conversation.as_deref(),
)
.map_err(|error| LibsyError::client_call(model.clone(), error))?;
}
LlmResponse::Agg(agg)
}
Expand Down Expand Up @@ -757,6 +770,7 @@ impl ClientRouter {
&model,
store,
input,
conversation.as_deref(),
)?;
}
return Ok(None);
Expand Down Expand Up @@ -842,6 +856,7 @@ impl ClientRouter {
model: &ModelId,
store: bool,
input: &CanonicalInput,
conversation: Option<&str>,
) -> std::result::Result<(), LlmClientError> {
let response_id = response.id.as_deref().filter(|_| store);
if response_id.is_none() {
Expand All @@ -856,11 +871,12 @@ impl ClientRouter {
input.parent.clone(),
Arc::from(segment),
));
let result =
self.inner
.state_owners
.lock()
.remember(response_id, None, model, Some(history));
let result = self.inner.state_owners.lock().remember(
response_id,
conversation,
Comment thread
coderabbitai[bot] marked this conversation as resolved.
model,
Some(history),
);
if let Err(error) = result {
if matches!(error, LlmClientError::ResponseStateLimitExceeded { .. }) {
tracing::warn!(%error, "cross-format Responses state capacity reached; history was not retained");
Expand Down Expand Up @@ -1283,6 +1299,191 @@ mod tests {
Ok(())
}

#[tokio::test]
async fn cross_format_conversation_lookup_finds_canonical_history() -> Result<()> {
let server = MockServer::start().await;
Mock::given(method("POST"))
.respond_with(|_request: &wiremock::Request| {
ResponseTemplate::new(200).set_body_json(json!({
"id": "chatcmpl_seed", "object": "chat.completion", "model": "weak",
"choices": [{
"index": 0,
"message": {"role": "assistant", "content": "seed answer"},
"finish_reason": "stop"
}],
"usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}
}))
})
.mount(&server)
.await;

let client: Arc<dyn RoutedLlmClient> = Arc::new(
TranslatingLlmClient::new(&[ModelConfig::new(
"weak",
Backend::OpenAiChat(HttpBackendConfig {
base_url: server.uri(),
api_key: None,
forward_auth: false,
extra_headers: BTreeMap::new(),
extra_body: BTreeMap::new(),
reasoning_effort: None,
max_retries: 0,
timeout: None,
}),
None,
)])
.map_err(|error| LibsyError::external("building test client", error))?,
);
let clients = ClientRouter::new(HashMap::from([(ModelId::from("weak"), client)]));
let models = to_category_map(&["weak"]);

let mut seed = Request {
llm_request: switchyard_translation::decode_request(
WireFormat::OpenAiResponses,
&json!({"model": "route", "input": "SEED", "store": true}),
)
.map_err(|error| LibsyError::external("decoding seed request", error))?,
raw_request: None,
metadata: None,
};
seed.llm_request
.extensions
.fields
.insert("conversation".to_string(), json!("conv_1"));

let (_, response) = run(
Arc::new(switchyard_libsy::Passthrough),
clients.clone(),
seed,
models,
None,
)
.await?;
response
.llm_response
.into_agg()
.await
.map_err(|error| LibsyError::client_call("weak", error))?;

let mut probe = Request {
llm_request: switchyard_translation::decode_request(
WireFormat::OpenAiResponses,
&json!({"model": "route", "input": "RECALL"}),
)
.map_err(|error| LibsyError::external("decoding continuation request", error))?,
raw_request: None,
metadata: None,
};
probe
.llm_request
.extensions
.fields
.insert("conversation".to_string(), json!("conv_1"));

let owner = clients
.stored_state_owner(&probe)
.expect("conversation id finds the recorded canonical history");
assert_eq!(owner.trigger, "conversation");
let history = owner.history.expect("materialized canonical history");
assert_eq!(history.segment.len(), 2);
Ok(())
}

#[tokio::test]
async fn conversation_lookup_returns_latest_materialized_history() -> Result<()> {
let server = MockServer::start().await;
Mock::given(method("POST"))
.respond_with(|_request: &wiremock::Request| {
ResponseTemplate::new(200).set_body_json(json!({
"id": "chatcmpl_seed", "object": "chat.completion", "model": "weak",
"choices": [{
"index": 0,
"message": {"role": "assistant", "content": "seed answer"},
"finish_reason": "stop"
}],
"usage": {"prompt_tokens": 1, "completion_tokens": 1, "total_tokens": 2}
}))
})
.mount(&server)
.await;

let client: Arc<dyn RoutedLlmClient> = Arc::new(
TranslatingLlmClient::new(&[ModelConfig::new(
"weak",
Backend::OpenAiChat(HttpBackendConfig {
base_url: server.uri(),
api_key: None,
forward_auth: false,
extra_headers: BTreeMap::new(),
extra_body: BTreeMap::new(),
reasoning_effort: None,
max_retries: 0,
timeout: None,
}),
None,
)])
.map_err(|error| LibsyError::external("building test client", error))?,
);
let clients = ClientRouter::new(HashMap::from([(ModelId::from("weak"), client)]));
let models = to_category_map(&["weak"]);

for input in ["FIRST", "SECOND"] {
let mut request = Request {
llm_request: switchyard_translation::decode_request(
WireFormat::OpenAiResponses,
&json!({"model": "route", "input": input, "store": true}),
)
.map_err(|error| LibsyError::external("decoding request", error))?,
raw_request: None,
metadata: None,
};
request
.llm_request
.extensions
.fields
.insert("conversation".to_string(), json!("conv_1"));

let (_, response) = run(
Arc::new(switchyard_libsy::Passthrough),
clients.clone(),
request,
models.clone(),
None,
)
.await?;
response
.llm_response
.into_agg()
.await
.map_err(|error| LibsyError::client_call("weak", error))?;
}

let mut probe = Request {
llm_request: switchyard_translation::decode_request(
WireFormat::OpenAiResponses,
&json!({"model": "route", "input": "RECALL"}),
)
.map_err(|error| LibsyError::external("decoding continuation request", error))?,
raw_request: None,
metadata: None,
};
probe
.llm_request
.extensions
.fields
.insert("conversation".to_string(), json!("conv_1"));

let owner = clients
.stored_state_owner(&probe)
.expect("conversation id finds the recorded canonical history");
assert_eq!(owner.trigger, "conversation");
let history = owner.history.expect("materialized canonical history");
// The second response stored the full four-message history; a stale
// first-writer entry would only hold the first two messages.
assert_eq!(history.segment.len(), 4);
Ok(())
}

#[tokio::test]
async fn responses_stored_tool_continuation_materializes_for_anthropic() -> Result<()> {
let server = MockServer::start().await;
Expand Down
Loading