schaefera/doubleo7
1
0
Fork 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

Add document embedding and retrieval via a dedicated embedding model #10

Merged
schaefera merged 10 commits from worktree-deep-research-max-turns-report into master 2026-08-19 15:37:33 +00:00
Conversation 0 Commits 10 Files changed 12 +82 -72
7 changed files with 82 additions and 72 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files
Showing only changes of commit 53b4f572df - Show all commits

49
Cargo.lock generated
View file

@ -5495,9 +5495,9 @@ checksum = "47b34b781b31e5d73e9fbc8689c70551fd1ade9a19e3e28cfec8580a79290cc4"
[[package]]
name = "rig"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "2ce03971e6115d30ef53fb3244d06718a4e62bbde82c103065600c09459b989a"
checksum = "b989f4060e5f3c1ea4b8b6999f6759949d820704286cc6976b7a688cc0133029"
dependencies = [
"rig-agent",
"rig-core",
@ -5510,9 +5510,9 @@ dependencies = [
[[package]]
name = "rig-agent"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "2b0796bbf47d7b76670401aac975bc619cf7fba3482b22dfe14992edaa9c2e04"
checksum = "b66477c0d0a786358d078e7e3541a73089dd9009100d6e2b3fe3b0d730e8550f"
dependencies = [
"async-stream",
"fastrand",
@ -5532,9 +5532,9 @@ dependencies = [
[[package]]
name = "rig-core"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "35f5520515ae8f6851adcbc6fde9eea8e96f657418c062e16c82cd81cce44e8e"
checksum = "432d83e0facf16749f91fe729cbffca84437e8062d2f4e92f4f12e903693922d"
dependencies = [
"as-any",
"async-stream",
@ -5556,6 +5556,7 @@ dependencies = [
"schemars 1.2.2",
"serde",
"serde_json",
"sha2",
"thiserror",
"tokio",
"tokio-tungstenite",
@ -5566,9 +5567,9 @@ dependencies = [
[[package]]
name = "rig-derive"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "eb868fcebdf3ba425e3afad2e4926bb6d9e1188a856843b00bcee2e15c07424f"
checksum = "de0a33f1bac45f16e50146c248bcbbfaa44518c7252d274e972c7f4ad71aaba7"
dependencies = [
"convert_case",
"proc-macro-crate",
@ -5579,9 +5580,9 @@ dependencies = [
[[package]]
name = "rig-fastembed"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "69ba93356242940918fac76a4b1cfe78b964f757e5cb65bb3feebd597635cfb0"
checksum = "f1dff99f9bea13108bb14df5431b6fc0738beb87f1fe5ebfcf26ae9fe54a188d"
dependencies = [
"fastembed",
"rig-core",
@ -5593,9 +5594,9 @@ dependencies = [
[[package]]
name = "rig-helixdb"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d1d40ae8fa89af33793626b898ec527e246183ef8b4b5095afee5f6a65cbbb1a"
checksum = "b5fa2b188a1e9e65d29c34738c97465f0822c57540d6d45fd4f04dad906c966a"
dependencies = [
"reqwest 0.13.4",
"rig-core",
@ -5606,12 +5607,11 @@ dependencies = [
[[package]]
name = "rig-lancedb"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f703b9e081ee4d77e60953598a5071a5cff4b8b702a2f06d251310950b0dae7c"
checksum = "c5fe3555202c5bd4e30948414c2e234fddd9260d4520046733fcf0d58718f3a1"
dependencies = [
"arrow-array",
"deranged",
"futures",
"lancedb",
"rig-core",
@ -5621,9 +5621,9 @@ dependencies = [
[[package]]
name = "rig-milvus"
version = "0.41.0"
version = "0.42.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "76662eb7f75de4e54b5f71e616430dc7f75d96d32f109d9435ad426b511eb565"
checksum = "af9c7b7ff7ed9c16da983a21f1e81f425c9805355427a92f4d2d97ba1fe2aa12"
dependencies = [
"reqwest 0.13.4",
"rig-core",
@ -6709,9 +6709,9 @@ dependencies = [
[[package]]
name = "tokio-tungstenite"
version = "0.28.0"
version = "0.29.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d25a406cddcc431a75d3d9afc6a7c0f7428d4891dd973e4d54c56b46127bf857"
checksum = "8f72a05e828585856dacd553fba484c242c46e391fb0e58917c942ee9202915c"
dependencies = [
"futures-util",
"log",
@ -6898,9 +6898,9 @@ checksum = "e421abadd41a4225275504ea4d6566923418b7f05506fbc9c0fe86ba7396114b"
[[package]]
name = "tungstenite"
version = "0.28.0"
version = "0.29.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "8628dcc84e5a09eb3d8423d6cb682965dea9133204e8fb3efee74c2a0c259442"
checksum = "6c01152af293afb9c7c2a57e4b559c5620b421f6d133261c60dd2d0cdb38e6b8"
dependencies = [
"bytes",
"data-encoding",
@ -6912,7 +6912,6 @@ dependencies = [
"rustls-pki-types",
"sha1",
"thiserror",
"utf-8",
]
[[package]]
@ -7033,12 +7032,6 @@ dependencies = [
"serde",
]
[[package]]
name = "utf-8"
version = "0.7.6"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "09cc8ee72d2a9becf2f2febe0205bbed8fc6615b7cb429ad062dc7b7ddd036a9"
[[package]]
name = "utf8-ranges"
version = "1.0.5"

2
Cargo.toml
View file

@ -16,7 +16,7 @@ clap = { version = "4", features = ["derive"] }
futures = "0.3"
indicatif = "0.18.6"
reqwest = { version = "0.13.4", features = ["query", "json"] }
rig = "0.41.0"
rig = "0.42.0"
schemars = "1"
scraper = "0.27"
serde = { version = "1.0.229", features = ["derive"] }

30
src/history.rs
View file

@ -149,11 +149,11 @@ mod tests {
fn assistant_text_ignores_tool_calls() {
let message = Message::Assistant {
id: None,
content: rig::OneOrMany::one(AssistantContent::tool_call(
content: vec![AssistantContent::tool_call(
"call-1",
"search_web",
serde_json::json!({ "query": "test" }),
)),
serde_json::json!({ "query": "test"}),
)],
};
assert!(extract_assistant_text(&message).is_empty());
}
@ -166,13 +166,13 @@ mod tests {
#[test]
fn tool_result_text_extracts_and_truncates() {
let short = Message::tool_result("call-1", "short result");
let short = Message::tool_result("call-1", "call-1-name", "short result");
assert_eq!(
tool_result_text(&short, 100),
vec!["short result".to_string()]
);
let long = Message::tool_result("call-2", "0123456789");
let long = Message::tool_result("call-2", "call-2-result", "0123456789");
assert_eq!(
tool_result_text(&long, 5),
vec!["01234 ...[truncated]".to_string()]
@ -191,11 +191,11 @@ mod tests {
Message::system("preamble"),
Message::Assistant {
id: None,
content: rig::OneOrMany::one(AssistantContent::tool_call(
content: vec![AssistantContent::tool_call(
"call-1",
"search_web",
serde_json::json!({ "query": "test" }),
)),
)],
},
];
@ -209,9 +209,9 @@ mod tests {
fn partial_findings_from_history_collects_assistant_text_and_tool_results_in_order() {
let history = vec![
Message::assistant("Checking sources..."),
Message::tool_result("call-1", "Result A [1]"),
Message::tool_result("call-1", "call-1-name", "Result A [1]"),
Message::assistant("Cross-checking..."),
Message::tool_result("call-2", "Result B [2]"),
Message::tool_result("call-2", "call-2-name", "Result B [2]"),
];
assert_eq!(
@ -233,11 +233,11 @@ mod tests {
fn transcript_lines_renders_tool_calls_with_their_arguments() {
let message = Message::Assistant {
id: None,
content: rig::OneOrMany::one(AssistantContent::tool_call(
content: vec![AssistantContent::tool_call(
"call-1",
"search_web",
serde_json::json!({ "query": "uruguay senior engineers" }),
)),
)],
};
assert_eq!(
transcript_lines(&message),
@ -247,7 +247,7 @@ mod tests {
#[test]
fn transcript_lines_prefixes_and_truncates_tool_results() {
let short = Message::tool_result("call-1", "found via search [1]");
let short = Message::tool_result("call-1", "call-1-name", "found via search [1]");
assert_eq!(
transcript_lines(&short),
vec!["Result: found via search [1]".to_string()]
@ -264,13 +264,13 @@ mod tests {
let history = vec![
Message::Assistant {
id: None,
content: rig::OneOrMany::one(AssistantContent::tool_call(
content: vec![AssistantContent::tool_call(
"call-1",
"search_web",
serde_json::json!({ "query": "test" }),
)),
)],
},
Message::tool_result("call-1", "1. Example\n https://example.com\n snippet"),
Message::tool_result("call-1", "call-1-name", "1. Example\n https://example.com\n snippet"),
Message::assistant("that source looks solid"),
];

9
src/research.rs
View file

@ -43,20 +43,17 @@ pub(crate) async fn research(
findings = gathered.findings;
incomplete = gathered.incomplete;
// The researcher ran out of turns mid-investigation rather than
// concluding on its own — another round would just repeat the same
// dead end, so stop and write up whatever was gathered.
// Prevent repeat iterations where researcher hits same dead end.
if incomplete {
tracing::info!(round, "researcher exhausted its turn budget");
break;
}
let review = review::review_findings(&client, topic, &findings, show_progress).await?;
let approved = review.approved;
tracing::info!(round, approved, "review verdict");
tracing::info!(round, review.approved, "review verdict");
if approved || round == MAX_RESEARCH_ROUNDS {
if review.approved || round == MAX_RESEARCH_ROUNDS {
break;
}

47
src/review.rs
View file

@ -1,4 +1,5 @@
use rig::client::AgentClientExt;
use rig::completion::TypedPrompt;
use rig::providers::ollama;
use rig::schemars::JsonSchema;
use serde::{Deserialize, Serialize};
@ -25,10 +26,18 @@ pub(crate) struct Review {
pub(crate) gaps: String,
}
/// Uses rig's typed extractor — a forced tool call into a `submit(Review)`
/// schema — rather than parsing free-text output, so the verdict and its
/// two feedback fields always come back structured instead of relying on
/// scanning prose for a trailing yes/no.
/// Retries for a review that fails to come back as valid structured output
/// (e.g. a transient network error), on top of the initial attempt.
const REVIEW_RETRIES: usize = 2;
/// Uses rig's typed-prompt API — `Native` structured output constraining the
/// model's own reply to the `Review` schema — rather than a forced tool call
/// or parsing free-text output. A forced tool call needs `tool_choice`
/// support, which Ollama doesn't have: the reviewer's small local model would
/// just answer in prose and never call the tool, so extraction only ever
/// exhausted its retries and errored out. `Native` mode instead uses Ollama's
/// own `format` JSON-schema constraint, which is honored regardless of
/// tool-calling ability.
#[tracing::instrument(skip(client, findings), fields(gen_ai.agent.name = "reviewer"))]
pub(crate) async fn review_findings(
client: &ollama::Client,
@ -37,7 +46,7 @@ pub(crate) async fn review_findings(
show_progress: bool,
) -> anyhow::Result<Review> {
let reviewer = client
.extractor::<Review>(REVIEWER_MODEL)
.agent(REVIEWER_MODEL)
.preamble(
"You are a skeptical fact-checker reviewing another researcher's notes before they \
get turned into a report. Approve only if every conclusion in the findings is \
@ -47,20 +56,36 @@ pub(crate) async fn review_findings(
Always separate the solid, well-supported findings from the gaps so a follow-up \
research pass knows what to keep and what to dig into further.",
)
.retries(2)
.build();
let spinner = Spinner::start(
show_progress,
format!("{REVIEW_EMOJI} Reviewing findings..."),
);
let review = reviewer
.extract(format!(
"Topic: {topic}\n\nResearch findings to review:\n{findings}"
))
.await?;
let prompt = format!("Topic: {topic}\n\nResearch findings to review:\n{findings}");
let mut last_error = None;
let mut extracted_review = None;
for attempt in 0..=REVIEW_RETRIES {
match reviewer.prompt_typed(prompt.clone()).await {
Ok(r) => {
extracted_review = Some(r);
break;
}
Err(e) => {
tracing::warn!("Attempt {attempt} to extract JSON failed: {e:?}.");
last_error = Some(e);
}
}
}
drop(spinner);
let review: Review = extracted_review
.ok_or_else(|| last_error.expect("loop always sets last_error on failure"))?;
tracing::info!(approved = review.approved, gaps = %review.gaps, "review complete");
Ok(review)

16
src/stream.rs
View file

@ -18,14 +18,11 @@ use std::io::Write;
/// covering the gap between the prompt being sent and generation starting
/// (otherwise the terminal would go blank for however long that takes)
/// rather than being dropped by the caller before this is even called.
pub(crate) async fn write_text_stream<R>(
mut stream: impl Stream<Item = Result<MultiTurnStreamItem<R>, StreamingError>> + Unpin,
pub(crate) async fn write_text_stream(
mut stream: impl Stream<Item = Result<MultiTurnStreamItem, StreamingError>> + Unpin,
writer: &mut impl Write,
mut spinner: Spinner,
) -> anyhow::Result<String>
where
R: Clone,
{
) -> anyhow::Result<String> {
let mut text = String::new();
while let Some(chunk) = stream.next().await {
@ -50,10 +47,7 @@ mod tests {
use rig::completion::CompletionError;
use rig::message::Text;
#[derive(Clone)]
struct DummyResponse;
fn text_item(text: &str) -> Result<MultiTurnStreamItem<DummyResponse>, StreamingError> {
fn text_item(text: &str) -> Result<MultiTurnStreamItem, StreamingError> {
Ok(MultiTurnStreamItem::StreamAssistantItem(
StreamedAssistantContent::Text(Text::new(text)),
))
@ -76,7 +70,7 @@ mod tests {
#[tokio::test]
async fn ignores_non_text_items() {
let final_item = Ok(MultiTurnStreamItem::final_response(
rig::OneOrMany::one(rig::message::AssistantContent::text("ignored")),
vec![rig::message::AssistantContent::text("ignored")],
rig::completion::Usage::new(),
));
let items = vec![text_item("kept"), final_item];

1
src/tools.rs
View file

@ -44,6 +44,7 @@ pub(crate) async fn search_web(
) -> Result<String, ToolExecutionError> {
progress::set_activity(format!("{SEARCH_EMOJI} Searching: {query}"));
// TODO re-use same client object?
let response = reqwest::Client::new()
.get(format!("{}/search", searxng_base_url()))
.query(&[("q", query.as_str()), ("format", "json")])