Apps/ImageApi
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

feat(ai): USER_NAME env + shared summary prompt + test-bin knobs

Browse Source 
Introduces USER_NAME (default "Me") as the single source for the message
sender label and the first-person persona across daily summaries, SMS
context, insight generation, and chat. Eliminates the "Me:" transcript /
"what I did" ambiguity that confused smaller models, and unhardcodes
"Cameron" from prompt text + the knowledge-graph owner entity. Set
USER_NAME=Cameron in .env to preserve the existing owner entity row
(keyed on UNIQUE(name, entity_type)) — otherwise the next run creates
a fresh owner entity and orphans the existing facts/photo-links.

Also:
- search_messages redirect: when the model calls it with date/contact
  but no query, return a hint pointing at get_sms_messages instead of
  a bare missing-parameter error (prevents same-turn retry loops)
- sharpen search_messages vs get_sms_messages tool descriptions so
  content-vs-time-based intent is unambiguous
- extract build_daily_summary_prompt (+ DAILY_SUMMARY_MESSAGE_LIMIT,
  DAILY_SUMMARY_SYSTEM_PROMPT) shared by daily_summary_job and
  test_daily_summary binary — prompt tweaks now land in both
- EMBEDDING_MODEL const; fixes both insert sites that stored
  "mxbai-embed-large:335m" while generate_embeddings actually runs
  "nomic-embed-text:v1.5"
- test_daily_summary: add --num-ctx / --temperature / --top-p /
  --top-k / --min-p flags wired into OllamaClient setters, and print
  the configured knobs at the top of each run
- OllamaClient::generate now logs prompt/gen token counts and tok/s
  via log_chat_metrics (symmetric with chat_with_tools)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Cameron
2026-04-22 23:39:37 -04:00
parent e4a3536f87
commit 6831f50993
6 changed files with 226 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
src/ai/ollama.rs
View File

@@ -19,6 +19,11 @@ pub use crate::ai::llm_client::{ToolCall, ToolCallFunction, ToolFunction};
// Cache duration: 15 minutes
const CACHE_DURATION_SECS: u64 = 15 * 60;
/// Embedding model used across the app. Callers that persist a
/// `model_version` alongside an embedding should read this constant so the
/// stored label always matches what `generate_embeddings` actually ran.
pub const EMBEDDING_MODEL: &str = "nomic-embed-text:v1.5";
// Cached entry with timestamp
#[derive(Clone)]
struct CachedEntry<T> {
@@ -349,6 +354,12 @@ impl OllamaClient {
}
let result: OllamaResponse = response.json().await?;
log_chat_metrics(
result.prompt_eval_count,
result.prompt_eval_duration,
result.eval_count,
result.eval_duration,
);
Ok(result.response)
}
@@ -481,6 +492,7 @@ Capture the key moment or theme. Return ONLY the title, nothing else."#,
) -> Result<String> {
let location_str = location.unwrap_or("Unknown");
let sms_str = sms_summary.unwrap_or("No messages");
let user_name = crate::ai::user_display_name();
let prompt = if image_base64.is_some() {
if let Some(contact_name) = contact {
@@ -492,13 +504,14 @@ Location: {}
Person/Contact: {}
Messages: {}
Analyze the image and use specific details from both the visual content and the context above. The photo is from a folder for {}, so they are likely in or related to this photo. Mention people's names (especially {}), places, or activities if they appear in either the image or the context. Write in first person as Cameron with the tone of a journal entry. If limited information is available, keep it simple and factual based on what you see and know. If the location is unknown omit it"#,
Analyze the image and use specific details from both the visual content and the context above. The photo is from a folder for {}, so they are likely in or related to this photo. Mention people's names (especially {}), places, or activities if they appear in either the image or the context. Write in first person as {} with the tone of a journal entry. If limited information is available, keep it simple and factual based on what you see and know. If the location is unknown omit it"#,
date.format("%B %d, %Y"),
location_str,
contact_name,
sms_str,
contact_name,
contact_name
contact_name,
user_name
)
} else {
format!(
@@ -508,10 +521,11 @@ Date: {}
Location: {}
Messages: {}
Analyze the image and use specific details from both the visual content and the context above. Mention people's names, places, or activities if they appear in either the image or the context. Write in first person as Cameron with the tone of a journal entry. If limited information is available, keep it simple and factual based on what you see and know. If the location is unknown omit it"#,
Analyze the image and use specific details from both the visual content and the context above. Mention people's names, places, or activities if they appear in either the image or the context. Write in first person as {} with the tone of a journal entry. If limited information is available, keep it simple and factual based on what you see and know. If the location is unknown omit it"#,
date.format("%B %d, %Y"),
location_str,
sms_str
sms_str,
user_name
)
}
} else if let Some(contact_name) = contact {
@@ -523,13 +537,14 @@ Analyze the image and use specific details from both the visual content and the
Person/Contact: {}
Messages: {}
Use only the specific details provided above. The photo is from a folder for {}, so they are likely related to this moment. Mention people's names (especially {}), places, or activities if they appear in the context. Write in first person as Cameron with the tone of a journal entry. If limited information is available, keep it simple and factual. If the location is unknown omit it"#,
Use only the specific details provided above. The photo is from a folder for {}, so they are likely related to this moment. Mention people's names (especially {}), places, or activities if they appear in the context. Write in first person as {} with the tone of a journal entry. If limited information is available, keep it simple and factual. If the location is unknown omit it"#,
date.format("%B %d, %Y"),
location_str,
contact_name,
sms_str,
contact_name,
contact_name
contact_name,
user_name
)
} else {
format!(
@@ -539,10 +554,11 @@ Analyze the image and use specific details from both the visual content and the
Location: {}
Messages: {}
Use only the specific details provided above. Mention people's names, places, or activities if they appear in the context. Write in first person as Cameron with the tone of a journal entry. If limited information is available, keep it simple and factual. If the location is unknown omit it"#,
Use only the specific details provided above. Mention people's names, places, or activities if they appear in the context. Write in first person as {} with the tone of a journal entry. If limited information is available, keep it simple and factual. If the location is unknown omit it"#,
date.format("%B %d, %Y"),
location_str,
sms_str
sms_str,
user_name
)
};
@@ -892,7 +908,7 @@ Analyze the image and use specific details from both the visual content and the
/// Returns a vector of 768-dimensional vectors
/// This is much more efficient than calling generate_embedding multiple times
pub async fn generate_embeddings(&self, texts: &[&str]) -> Result<Vec<Vec<f32>>> {
let embedding_model = "nomic-embed-text:v1.5";
let embedding_model = EMBEDDING_MODEL;
log::debug!("=== Ollama Batch Embedding Request ===");
log::debug!("Model: {}", embedding_model);
@@ -1145,6 +1161,14 @@ struct OllamaStreamMessage {
#[derive(Deserialize)]
struct OllamaResponse {
response: String,
#[serde(default)]
prompt_eval_count: Option<i32>,
#[serde(default)]
prompt_eval_duration: Option<u64>,
#[serde(default)]
eval_count: Option<i32>,
#[serde(default)]
eval_duration: Option<u64>,
}
fn log_chat_metrics(