feat: support local Ollama for embeddings and fact-extraction LLM by imonroe · Pull Request #79 · imonroe/memserv

imonroe · 2026-07-08T17:52:10Z

Closes #51.

What

Adds first-class, opt-in config for running mem0's embedder and/or fact-extraction LLM against a local Ollama server — zero per-call cost and no content leaving the host — while keeping anthropic + openai as the default. As the issue notes, _build_config already abstracts the providers, so this is a config + wiring + docs change, not a re-architecture. No change to the single-Memory-per-process invariant.

Changes

app/config.py — new OLLAMA_BASE_URL setting (default http://localhost:11434). The existing _require_provider_keys validator already enforces a provider's key only when that provider is selected, so ollama needs no ANTHROPIC_API_KEY/OPENAI_API_KEY.
app/memory.py — _provider_config() is now provider-aware: cloud providers get the API key injected explicitly (as before), while an ollama provider gets mem0's ollama_base_url key (and embedding_dims for the embedder) and no key. Wired through _build_config() for both llm and embedder, so fully-local and mixed cloud/local setups both work.
.env.example — documented OLLAMA_BASE_URL + a commented-out fully-local provider block.
docker-compose.yml — commented-out optional bundled ollama service (+ ollama_data volume) for the self-contained stack, with model-pull notes and an optional GPU block.
docs/USER_GUIDE.md — new "Running fully local with Ollama" section and config-table rows, prominently restating the critical MEM0_EMBED_DIMS-must-match invariant (switching embed models — including cloud↔local — means dropping/recreating the Qdrant collection). docs/DEVELOPER_GUIDE.md — updated the _provider_config internals note.

Config keys (verified against mem0 2.0.2)

{"llm":      {"provider": "ollama", "config": {"model": "llama3.1:8b", "ollama_base_url": "http://localhost:11434"}},
 "embedder": {"provider": "ollama", "config": {"model": "nomic-embed-text", "ollama_base_url": "http://localhost:11434", "embedding_dims": 768}}}

Verification

ruff check clean; 241 tests pass (pytest -q) against the pinned mem0ai==2.0.2.
New tests: Ollama config building (fully-local + mixed local-LLM/cloud-embed), OLLAMA_BASE_URL default/override, and — critically — the Ollama config is validated against the real mem0.configs.base.MemoryConfig schema, so a future mem0 rename of ollama_base_url is caught in CI rather than at deploy time.
Default behavior is unchanged: with no Ollama vars set, the config is byte-for-byte the same anthropic+openai shape as before (existing test_build_config_shape still passes untouched).

🤖 Generated with Claude Code

Generated by Claude Code

Adds opt-in config for running mem0's embedder and/or LLM against a local Ollama server — zero per-call cost and no content leaving the host — while keeping anthropic+openai as the default. - config.py: new OLLAMA_BASE_URL setting (default http://localhost:11434). The existing provider-key validator already skips key checks for non anthropic/openai providers, so ollama needs no API key. - memory.py: _provider_config() is now provider-aware — cloud providers get the API key injected; an "ollama" provider gets mem0's ollama_base_url (and embedding_dims for the embedder) and no key. Wired through _build_config for both llm and embedder, so fully-local and mixed cloud/local setups work. - .env.example / docker-compose.yml: documented vars plus a commented-out optional bundled `ollama` service for the Compose stack. - USER_GUIDE: "Running fully local with Ollama" section + config-table rows, prominently restating the MEM0_EMBED_DIMS-must-match invariant (recreate the Qdrant collection when switching embed models). DEVELOPER_GUIDE: updated the _provider_config internals note. - Tests: ollama config building (fully-local + mixed), OLLAMA_BASE_URL default/override, and Ollama config validated against the real mem0 MemoryConfig schema so a key rename is caught in CI. Closes #51. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XQXUARh5h5hRFg67toenEX

Copilot

Pull request overview

Adds opt-in support for running mem0’s embedder and/or fact-extraction LLM against a local Ollama server, while preserving the existing Anthropic (LLM) + OpenAI (embedder) defaults. This is primarily provider-config wiring plus documentation and tests to ensure the generated mem0 config matches the pinned mem0 schema.

Changes:

Add OLLAMA_BASE_URL setting and wire ollama_base_url (+ embedding_dims for embeddings) into mem0 provider config.
Expand docs and .env.example to document fully-local and mixed cloud/local setups, plus a commented optional ollama Docker Compose service.
Add tests for Ollama config shapes and schema validation against mem0.configs.base.MemoryConfig.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
`app/config.py`	Adds `ollama_base_url` setting (default `http://localhost:11434`).
`app/memory.py`	Makes provider config provider-aware and wires Ollama base URL + embedding dims into mem0 config.
`tests/test_memory.py`	Adds tests for Ollama-only and mixed provider config shapes + validates against real mem0 schema.
`tests/test_config.py`	Adds test coverage for `OLLAMA_BASE_URL` default/override.
`docs/USER_GUIDE.md`	Documents Ollama usage, config reference updates, and the embed-dims/Qdrant invariant.
`docs/DEVELOPER_GUIDE.md`	Updates internal notes about `_provider_config()` behavior for cloud vs Ollama.
`docker-compose.yml`	Adds a commented-out optional `ollama` service and volume notes.
`.env.example`	Documents Ollama-related env vars and a fully-local example block.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Addresses PR review: _build_config previously hardcoded the Anthropic key into the LLM config and the OpenAI key into the embedder config regardless of the configured provider. With openai now documented as a valid MEM0_LLM_PROVIDER, an OpenAI LLM would have received the Anthropic key (and no OpenAI key), silently failing at first request. - memory.py: add _api_key_for(provider) so each provider gets its own key (anthropic->ANTHROPIC_API_KEY, openai->OPENAI_API_KEY, ollama/other->none). - config.py: _require_provider_keys now requires a provider's key when *either* role selects it, so openai-LLM + ollama-embed no longer passes validation without OPENAI_API_KEY. - Tests: openai-LLM key routing, and validation raises for a missing key when openai/anthropic is the LLM (or embed) provider. - Docs: config-table key requirements and the DEVELOPER_GUIDE validator note updated to reflect either-role enforcement. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XQXUARh5h5hRFg67toenEX

Copilot

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Addresses PR review follow-ups: - config.py: reword the _require_provider_keys comment — mem0's clients DO read keys from os.environ; the reason we inject is that a key loaded from .env via pydantic-settings never lands in os.environ. The old wording said the opposite and contradicted the note in app/memory.py. - tests: remove test_missing_anthropic_key_rejected_for_embed_provider. It implied Anthropic is a valid embed provider, but Anthropic has no embeddings API and the docs list only openai/ollama for embedding — the real key-backed cases (anthropic LLM, openai LLM or embed) stay covered. Also drop the same misleading os.environ phrasing from the openai-LLM test comment. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XQXUARh5h5hRFg67toenEX

imonroe requested a review from Copilot July 8, 2026 17:56

Copilot started reviewing on behalf of imonroe July 8, 2026 17:56 View session

Copilot AI reviewed Jul 8, 2026

View reviewed changes

Comment thread app/memory.py

Comment thread tests/test_config.py

imonroe requested a review from Copilot July 8, 2026 18:32

Copilot started reviewing on behalf of imonroe July 8, 2026 18:32 View session

Copilot AI reviewed Jul 8, 2026

View reviewed changes

Comment thread app/config.py Outdated

Comment thread docs/USER_GUIDE.md

imonroe merged commit a37353a into main Jul 9, 2026
1 check passed

Navigation Menu

feat: support local Ollama for embeddings and fact-extraction LLM - #79

feat: support local Ollama for embeddings and fact-extraction LLM#79
imonroe merged 3 commits into
mainfrom
claude/github-issue-51-ollama

imonroe commented Jul 8, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

imonroe commented Jul 8, 2026

What

Changes

Config keys (verified against mem0 2.0.2)

Verification

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants