Skip to content

Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors

Rust

fix(llm): finish Google embedding integration

AkitaOnRailsf0702571df9d
diff --git a/CHANGELOG.md b/CHANGELOG.md
index f9bd5616..9810800d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -34,6 +34,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
- Google Gemini LLM provider via `AI_MEMORY_LLM_PROVIDER=gemini`, with
`gemini-2.5-flash` as the default hosted Google model and `GEMINI_API_KEY`
/ `GOOGLE_API_KEY` support.
+- Google Gemini embeddings via `AI_MEMORY_EMBEDDING_PROVIDER=google` or
+ `gemini`, with `gemini-embedding-001` as the default embedding model and
+ `GEMINI_API_KEY` / `GOOGLE_API_KEY` support.
- README support matrix for operating systems, agent integrations, LLM
providers, and embedding providers.
diff --git a/README.md b/README.md
index 45acac1c..9f545349 100644
--- a/README.md
+++ b/README.md
@@ -31,7 +31,7 @@
| Claude Desktop | MCP-only | Uses `mcp-remote`; no lifecycle hooks. |
| OpenClaw | Supported | MCP config + native plugin lifecycle hooks. |
| LLM providers | Supported | Anthropic, OpenAI, Gemini, and OpenAI-compatible endpoints. |
-| Embedding providers | Supported | OpenAI and Voyage. |
+| Embedding providers | Supported | OpenAI, Voyage, and Google Gemini. |
## What it is
@@ -271,8 +271,8 @@ Recommended defaults:
| `openai-compat` | no default | OpenRouter, Ollama, vLLM, LM Studio, and other compatible endpoints. |
Embeddings are optional and separate from the LLM provider. Set
-`AI_MEMORY_EMBEDDING_PROVIDER=openai` or `voyage` when you want vector
-reranking in addition to FTS5 + graph-neighbor retrieval.
+`AI_MEMORY_EMBEDDING_PROVIDER=openai`, `voyage`, `google`, or `gemini` when
+you want vector reranking in addition to FTS5 + graph-neighbor retrieval.
See [`docs/install.md#llm-provider-tiers`](docs/install.md#llm-provider-tiers)
for env vars and Ollama/OpenRouter examples, and
diff --git a/crates/ai-memory-cli/src/commands/serve.rs b/crates/ai-memory-cli/src/commands/serve.rs
index e3872c09..0e2f3a4e 100644
--- a/crates/ai-memory-cli/src/commands/serve.rs
+++ b/crates/ai-memory-cli/src/commands/serve.rs
@@ -376,7 +376,7 @@ async fn run_embedding_backfill(
continue;
}
};
- let vec = match embedder.embed(&md.body).await {
+ let vec = match embedder.embed_document(&md.body).await {
Ok(vec) => vec,
Err(e) => {
failed += 1;
@@ -429,9 +429,14 @@ async fn configure_embedder(
info!("AI_MEMORY_EMBEDDING_PROVIDER unset; hybrid search disabled (FTS5-only)");
return Ok((wiki, None));
};
+ let embedder = build_embedder(cfg).context("building embedder from config")?;
let mismatch = store
.reader
- .embedding_meta_for_mismatch(cfg.provider.name().into(), cfg.model.clone(), cfg.dim)
+ .embedding_meta_for_mismatch(
+ embedder.provider().into(),
+ embedder.model().into(),
+ embedder.dim(),
+ )
.await?;
if !mismatch.is_empty() {
anyhow::bail!(
@@ -440,7 +445,6 @@ async fn configure_embedder(
mismatch
);
}
- let embedder = build_embedder(cfg).context("building embedder from config")?;
info!(
provider = embedder.provider(),
model = embedder.model(),
diff --git a/crates/ai-memory-cli/src/config.rs b/crates/ai-memory-cli/src/config.rs
index d19faa0e..668f6d27 100644
--- a/crates/ai-memory-cli/src/config.rs
+++ b/crates/ai-memory-cli/src/config.rs
@@ -375,13 +375,9 @@ impl Config {
.voyage_api_key
.clone()
.ok_or_else(|| LlmError::NotConfigured("VOYAGE_API_KEY".into()))?,
- EmbedderChoice::Google => self
- .runtime_env
- .gemini_api_key
- .clone()
- .ok_or_else(|| {
- LlmError::NotConfigured("GEMINI_API_KEY or GOOGLE_API_KEY".into())
- })?,
+ EmbedderChoice::Google => self.runtime_env.gemini_api_key.clone().ok_or_else(|| {
+ LlmError::NotConfigured("GEMINI_API_KEY or GOOGLE_API_KEY".into())
+ })?,
};
Ok(Some(EmbedderConfig {
provider,
@@ -515,4 +511,27 @@ mod tests {
assert!(!cfg.maintenance.enabled);
assert_eq!(cfg.maintenance.lint_interval_secs, 3600);
}
+
+ #[test]
+ fn gemini_embedding_provider_uses_google_defaults() {
+ let mut cfg = Config {
+ embedding_provider: Some("gemini".into()),
+ runtime_env: RuntimeEnv {
+ gemini_api_key: Some(SecretString::from("test-key")),
+ ..RuntimeEnv::default()
+ },
+ ..Config::default()
+ };
+
+ let embedder = cfg.embedder_config().unwrap().unwrap();
+ assert_eq!(embedder.provider, EmbedderChoice::Google);
+ assert_eq!(embedder.model, ai_memory_llm::GOOGLE_DEFAULT_EMBED_MODEL);
+ assert_eq!(embedder.dim, 768);
+
+ cfg.embedding_provider = Some("google".into());
+ assert_eq!(
+ cfg.embedder_config().unwrap().unwrap().provider,
+ EmbedderChoice::Google
+ );
+ }
}
diff --git a/crates/ai-memory-consolidate/tests/recall_eval.rs b/crates/ai-memory-consolidate/tests/recall_eval.rs
index 229940c2..f98b6159 100644
--- a/crates/ai-memory-consolidate/tests/recall_eval.rs
+++ b/crates/ai-memory-consolidate/tests/recall_eval.rs
@@ -306,7 +306,7 @@ async fn measure_recall(
let mut hits = 0_usize;
for (query, expected) in PROBES {
let results = if let Some(emb) = &embedder {
- let qv = emb.embed(query).await.expect("embed query");
+ let qv = emb.embed_query(query).await.expect("embed query");
store
.reader
.hybrid_search(
diff --git a/crates/ai-memory-llm/src/factory.rs b/crates/ai-memory-llm/src/factory.rs
index bab21bba..0fc9a141 100644
--- a/crates/ai-memory-llm/src/factory.rs
+++ b/crates/ai-memory-llm/src/factory.rs
@@ -12,8 +12,8 @@ use crate::GeminiProvider;
use crate::OpenAiCompatProvider;
use crate::OpenAiProvider;
use crate::embedding::{Embedder, OpenAiEmbedder, VoyageEmbedder};
-use crate::google::GoogleEmbedder;
use crate::error::{LlmError, LlmResult};
+use crate::google::GoogleEmbedder;
use crate::provider::LlmProvider;
/// Four providers ship in v1.
diff --git a/crates/ai-memory-llm/src/google.rs b/crates/ai-memory-llm/src/google.rs
index 1be16914..61473e59 100644
--- a/crates/ai-memory-llm/src/google.rs
+++ b/crates/ai-memory-llm/src/google.rs
@@ -79,11 +79,7 @@ impl GoogleEmbedder {
content: GeminiContent {
parts: vec![GeminiPart { text: &prepared }],
},
- task_type: if self.embedding_v2 {
- None
- } else {
- task_type
- },
+ task_type: if self.embedding_v2 { None } else { task_type },
output_dimensionality: Some(self.dim),
};
@@ -121,7 +117,10 @@ struct GeminiEmbedRequest<'a> {
content: GeminiContent<'a>,
#[serde(skip_serializing_if = "Option::is_none", rename = "taskType")]
task_type: Option<&'a str>,
- #[serde(skip_serializing_if = "Option::is_none", rename = "outputDimensionality")]
+ #[serde(
+ skip_serializing_if = "Option::is_none",
+ rename = "outputDimensionality"
+ )]
output_dimensionality: Option<u32>,
}
@@ -164,8 +163,7 @@ impl Embedder for GoogleEmbedder {
}
async fn embed_document(&self, text: &str) -> LlmResult<Vec<f32>> {
- self.embed_with_task(text, Some("RETRIEVAL_DOCUMENT"))
- .await
+ self.embed_with_task(text, Some("RETRIEVAL_DOCUMENT")).await
}
async fn embed_query(&self, text: &str) -> LlmResult<Vec<f32>> {
diff --git a/crates/ai-memory-llm/src/lib.rs b/crates/ai-memory-llm/src/lib.rs
index b1474e61..6df853f9 100644
--- a/crates/ai-memory-llm/src/lib.rs
+++ b/crates/ai-memory-llm/src/lib.rs
@@ -43,7 +43,7 @@ pub use factory::{
default_embedding_dim,
};
pub use gemini::GeminiProvider;
-pub use google::{GoogleEmbedder, DEFAULT_MODEL as GOOGLE_DEFAULT_EMBED_MODEL};
+pub use google::{DEFAULT_MODEL as GOOGLE_DEFAULT_EMBED_MODEL, GoogleEmbedder};
pub use openai::OpenAiProvider;
pub use openai_compat::OpenAiCompatProvider;
pub use provider::{LlmProvider, complete_structured};
diff --git a/crates/ai-memory-mcp/src/admin.rs b/crates/ai-memory-mcp/src/admin.rs
index dfd00487..24d3e400 100644
--- a/crates/ai-memory-mcp/src/admin.rs
+++ b/crates/ai-memory-mcp/src/admin.rs
@@ -962,7 +962,7 @@ async fn handle_embed(
continue;
}
};
- let vec = match embedder.embed(&md.body).await {
+ let vec = match embedder.embed_document(&md.body).await {
Ok(v) => v,
Err(e) => {
warn!(path = %cand.path, error = %e, "embed: provider call failed");
diff --git a/crates/ai-memory-mcp/src/server.rs b/crates/ai-memory-mcp/src/server.rs
index d4e38936..f54c942c 100644
--- a/crates/ai-memory-mcp/src/server.rs
+++ b/crates/ai-memory-mcp/src/server.rs
@@ -410,7 +410,7 @@ impl AiMemoryServer {
// instead of erroring the whole query — losing semantic
// re-ranking is acceptable, denying the user any search isn't.
let hits = if let Some(embedder) = &self.embedder {
- match embedder.embed(&args.query).await {
+ match embedder.embed_query(&args.query).await {
Ok(qv) => {
self.reader
.hybrid_search(
diff --git a/crates/ai-memory-wiki/src/wiki.rs b/crates/ai-memory-wiki/src/wiki.rs
index 5dbe4932..eabd7d2b 100644
--- a/crates/ai-memory-wiki/src/wiki.rs
+++ b/crates/ai-memory-wiki/src/wiki.rs
@@ -348,7 +348,7 @@ impl Wiki {
// transaction" (basic-memory #763 lesson): no fire-and-forget
// background embedding.
if let Some(embedder) = &self.embedder {
- match embedder.embed(&body).await {
+ match embedder.embed_document(&body).await {
Ok(vec) => {
let bytes = f32_vec_to_bytes(&vec);
self.writer
diff --git a/docs/ARCHITECTURE.md b/docs/ARCHITECTURE.md
index 28781217..46e3f694 100644
--- a/docs/ARCHITECTURE.md
+++ b/docs/ARCHITECTURE.md
@@ -269,10 +269,10 @@ AI_MEMORY_LLM_BASE_URL for openai-compat (Ollama, vLLM)
**Embedder env** (opt-in):
```
-AI_MEMORY_EMBEDDING_PROVIDER openai | voyage
-AI_MEMORY_EMBEDDING_MODEL e.g. text-embedding-3-small
-AI_MEMORY_EMBEDDING_DIM 1536
-OPENAI_API_KEY / VOYAGE_API_KEY
+AI_MEMORY_EMBEDDING_PROVIDER openai | voyage | google | gemini
+AI_MEMORY_EMBEDDING_MODEL e.g. text-embedding-3-small, gemini-embedding-001
+AI_MEMORY_EMBEDDING_DIM 1536 (OpenAI), 1024 (Voyage), 768 (Google)
+OPENAI_API_KEY / VOYAGE_API_KEY / GEMINI_API_KEY / GOOGLE_API_KEY
```
## Future work
diff --git a/docs/install.md b/docs/install.md
index 272fb4b4..010ad208 100644
--- a/docs/install.md
+++ b/docs/install.md
@@ -349,6 +349,7 @@ If you set only the provider, ai-memory picks a sensible default:
| `AI_MEMORY_LLM_PROVIDER=gemini` | `gemini-2.5-flash` | Google's hosted option with a generous free tier. ai-memory disables Gemini 2.5 Flash's default dynamic thinking so hidden thought tokens do not truncate strict JSON. Set `GEMINI_API_KEY` (or `GOOGLE_API_KEY`). |
| `AI_MEMORY_EMBEDDING_PROVIDER=openai` | `text-embedding-3-small` (1536-dim) | 5× cheaper than `-3-large` with marginal recall loss. |
| `AI_MEMORY_EMBEDDING_PROVIDER=voyage` | `voyage-3` (1024-dim) | Voyage's current general-purpose recommendation. |
+| `AI_MEMORY_EMBEDDING_PROVIDER=google` / `gemini` | `gemini-embedding-001` (768-dim) | Google-hosted embeddings via `embedContent`. Set `GEMINI_API_KEY` (or `GOOGLE_API_KEY`). |
> **What we don't recommend:** reasoning-mode models (Claude with extended
> thinking, GPT-o3, Gemini "thinking" variants) — they burn token budget on