Codebase Retrieval Not Working After Successful Indexing

Before submitting your bug report

• I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• I'm not able to find a related conversation on GitHub discussions that reports the same bug
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

Environment
Continue Extension: 1.2.22
VS Code: 1.123.0
OS: Ubuntu Linux 24.04.4 LTS
Local LLM Provider: Ollama
Chat Model: qwen2.5-coder:14b
Autocomplete Model: qwen2.5-coder:7b
Embedding Model: nomic-embed-text

Description

Continue reports that codebase indexing completed successfully, but the agent is unable to retrieve actual files from the workspace.

When asked to inspect the codebase, locate files, or review the project structure, the model hallucinates file paths and project structures that do not exist instead of retrieving information from the indexed workspace.

Additionally, indexing warnings were observed:

Embedding: Failed to generate embeddings for 357 chunks with provider:
_TransformersJsEmbeddingsProvider::all-MiniLM-L6-v2

Error: no available backend found.
ENOENT: no such file or directory,
ort-wasm-simd-threaded.wasm

and:

Full text search: SQLITE_ERROR: no such table: chunks

Although the indexing warnings later disappeared after re-indexing, codebase retrieval still does not appear to function correctly.

To reproduce

Steps to Reproduce
Install Continue 1.2.22 on VS Code 1.123.0.
Configure Ollama with:
qwen2.5-coder:14b
qwen2.5-coder:7b
nomic-embed-text
Open an existing project workspace.
Run:
Continue: Docs Index
Continue: Codebase Force Re-Index
Wait until indexing reports completion.
Ask the agent:
Find the file that defines the Usuario model.

Return only:

• file path
• class name

or:

List 20 actual files from the indexed workspace.
Expected Behavior

The agent should retrieve and reference actual files from the workspace.

For example, in my project it should return paths such as:

backend/app/models/usuario.py
backend/app/main.py
backend/app/core/config.py
backend/tests/test_email_verification.py

and should identify the actual Usuario model definition.

Actual Behavior

The agent returns hallucinated file paths and structures such as:

src/models/user.py
src/app.py
src/database/init.py

which do not exist in the workspace.

When asked to list actual files, the agent either invents files or states that it cannot access them despite indexing reporting success.

Additional Observations
Ollama integration itself works correctly.
Local models are being used successfully.
GPU utilization confirms that requests are executed locally.
Chat, edit, and apply functionality work as expected.
The issue appears specifically related to workspace retrieval / indexing.
The "Add more context providers" link opened a documentation page that returned "Page not found".
It is difficult to determine whether the agent is actually using indexed workspace data or only responding from model knowledge.

Log output

Reproduce the failure in Continue 1.2.22 on VS Code with Ollama and the listed models, then inspect the indexing and workspace-retrieval logs, especially the embedding backend error and the SQLite "no such table: chunks" error. Verify whether re-indexing creates usable workspace data by asking for the listed files and the Usuario model; done means the agent returns actual indexed paths and the correct class rather than hallucinated ones.