[Bug] How to use a local embedding model ?

[warlordattack]

Hello,

I try to use honcho in local homelab, with hermes and openclaw, everything is local,
no paid models services or subscriptions,
minimum use of internet, all data is local.

The goal is to survive zombi apocalypse,
and have minimal use of internet (only if no other choice).
Data and services must stay local.
For example i use self hosted synapse/element/matrix chat instead of Telegram,
local Searxng (search engine) that search into local Zimi and Kiwix (local selfhosted Wikipedia and other ZIM files). Instead of Tailscale i use self hosted Headscale, local Rusdesk for RDP, etc ...

All my models are running in LM Studio on a gaming PC + RTX 4090 (+ other LM Studio instances distributed in lan).
My homelab is an old gaming PC with RTX 3070 and proxmox.
Proxmox runs Truenas VM, Home assistant VM, and linux Mint VM with GPU passthrough.
The Mint VM has docker with 100 containers, including AI containers, some of them use GPU (from Mint).
Hermes and Openclaw are installed directly inside Mint VM (i tryed docker versions and had many problems, so installed them inside linux Mint via bash one liner (no docker), and they work better).

Now i try to configure Honcho with both Openclaw and Hermes (the idea is they will share same memory through Honcho, and run in parallel using different LM Studio instances).
Honcho will be docker container, will use existing redis/postgres(pgvector18) docker containers,
and also LM Studio witch serve models from lan gaming PC (http://${PROART_IP}:1234/v1).

Everything is configured and running but there is two problems :
1/
Looks like there is no way to configure an LM Studio embedding model compatible with Honcho, in relation with this :
( Embedding dimensions (default: 1536 for OpenAI text-embedding-3-small)
VECTOR_STORE_DIMENSIONS=768 )
Honcho always sends "text-embedding-3-small" to LM Studio and forces 1536 dimensions instead of 768 ?
And i did not find any LM Studio embedding model with 1536 dimensions compatible with Honcho.
So Hermes and Openclaw crash due to embedding problem with Honcho.
2/
LM Studio lazy load models when recieve request, but this does not work with Honcho embedding, i need to load an embedding model manually (perhaps related to 1/).
This is not optimal, because you need to do manual action in LM Studio, and i have multiple instances.

Second env file (i have main env file with variables for all containers) :

# models : Lm Studio : honcho
MD_LMS_DIALECTIC=hermes-3-llama-3.1-8b      
MD_LMS_DERIVER=llama-3.2-3b-instruct       
MD_LMS_EMBEDDING=gte-qwen2-1.5b-instruct   
MD_LMS_SUMMARY=llama-3.2-3b-instruct        
...

Honcho compose :

# https://github.com/plastic-labs/honcho/blob/main/docker-compose.yml.example
# https://github.com/elkimek/honcho-self-hosted/tree/main

services:

  srv_honcho_api:
    extends:
      file: ../_common/all.yml
      service: all

    image: ${C_HONA}:latest
    build:
      context: ../../git/honcho
      dockerfile: Dockerfile

    env_file:
      - ./.env

    entrypoint: ["sh", "docker/entrypoint.sh"]

    container_name: ${C_HONA}
    ports:
      - ${C_HONA_8000}:8000 # 8800
    environment:
      # postgres + redis
      - DB_CONNECTION_URI=postgresql+psycopg://${C_HON_PG_USER}:${C_HON_PG_PWD}@${C_PG18v_HOST}:5432/${C_HON_PG_DB}
      - CACHE_URL=${C_RED_URL}?suppress=true
      # --- LM Studio ---
      - LLM_OPENAI_COMPATIBLE_BASE_URL=http://${PROART_IP}:1234/v1
      # --- Dialectic Levels (custom provider + model) ---
      - DIALECTIC_LEVELS__minimal__MODEL=${MD_LMS_DERIVER} 
      - DIALECTIC_LEVELS__low__MODEL=${MD_LMS_DERIVER}
      - DIALECTIC_LEVELS__medium__MODEL=${MD_LMS_DIALECTIC}
      - DIALECTIC_LEVELS__high__MODEL=${MD_LMS_DIALECTIC}
      - DIALECTIC_LEVELS__max__MODEL=${MD_LMS_DIALECTIC}
      # --- Embeddings ---
      - LLM_EMBEDDING_BASE_URL=http://${PROART_IP}:1234/v1
      - LLM_EMBEDDING_MODEL=${MD_LMS_EMBEDDING} 
      # --- Summary ---
      - SUMMARY_MODEL=${MD_LMS_SUMMARY}
      # --- Deriver ---
      - DERIVER_MODEL=${MD_LMS_DERIVER}

    hostname: ${C_HONA_HOST}  # hona
    networks:
      proxy:
        ipv4_address: ${NET_TFK}.${C_HONA_NUM} # 168
        mac_address: ${MAC_TFK}:${C_HONA_MAC}  # a8
    #  proxmox:
    #    ipv4_address: ${NET_PX}.${C_HONA_NUM}
    #    mac_address: ${MAC_PX}:${C_HONA_MAC} 
    #dns:
    #  - ${NET_PX_DNS}
      
    depends_on:
          - srv_honcho_deriver


  srv_honcho_deriver:
    extends:
      file: ../_common/all.yml
      service: all

    image: ${C_HONA}:latest
    build:
      context: ../../git/honcho
      dockerfile: Dockerfile

    env_file:
      - ./.env

    #command: python -m honcho.deriver
    entrypoint: ["/app/.venv/bin/python", "-m", "src.deriver"]

    container_name: ${C_HOND}
    environment:
      # postgres + redis
      - DB_CONNECTION_URI=postgresql+psycopg://${C_HON_PG_USER}:${C_HON_PG_PWD}@${C_PG18v_HOST}:5432/${C_HON_PG_DB}
      - CACHE_URL=${C_RED_URL}?suppress=true
      # --- LM Studio ---
      - LLM_OPENAI_COMPATIBLE_BASE_URL=http://${PROART_IP}:1234/v1
      # --- Dialectic Levels (custom provider + model) ---
      - DIALECTIC_LEVELS__minimal__MODEL=${MD_LMS_DERIVER} 
      - DIALECTIC_LEVELS__low__MODEL=${MD_LMS_DERIVER}
      - DIALECTIC_LEVELS__medium__MODEL=${MD_LMS_DIALECTIC}
      - DIALECTIC_LEVELS__high__MODEL=${MD_LMS_DIALECTIC}
      - DIALECTIC_LEVELS__max__MODEL=${MD_LMS_DIALECTIC}
      # --- Embeddings ---
      - LLM_EMBEDDING_BASE_URL=http://${PROART_IP}:1234/v1
      - LLM_EMBEDDING_MODEL=${MD_LMS_EMBEDDING}
      # --- Summary ---
      - SUMMARY_MODEL=${MD_LMS_SUMMARY}
      # --- Deriver ---
      - DERIVER_MODEL=${MD_LMS_DERIVER}

    hostname: ${C_HOND_HOST} # hond
    networks:
      proxy:
        ipv4_address: ${NET_TFK}.${C_HOND_NUM} # 169
        mac_address: ${MAC_TFK}:${C_HOND_MAC}  # a9
    #  proxmox:
    #    ipv4_address: ${NET_PX}.${C_HOND_NUM}
    #    mac_address: ${MAC_PX}:${C_HOND_MAC} 
    #dns:
    #  - ${NET_PX_DNS}

networks: 
  proxy: 
    external: true
  proxmox:
    external: true

# docker exec -it ai-honcho-api python -m honcho.manage upgrade
# curl http://192.168.50.54:1234/v1/models
# docker compose run --rm ai-honcho-api alembic upgrade head
# cd /Volumes/mb-docker/git
# git clone https://github.com/plastic-labs/honcho.git

Honcho .env ( env_file: - ./.env) :

# Honcho Environment Variables Template
# Copy this file to .env and fill in the appropriate values
#
# Required variables are marked with (REQUIRED)
# Optional variables have default values and can be left commented out

# =============================================================================
# Application Settings
# =============================================================================
LOG_LEVEL=INFO
# SESSION_OBSERVERS_LIMIT=10
# GET_CONTEXT_MAX_TOKENS=100000
# MAX_FILE_SIZE=5242880 # Bytes
# MAX_MESSAGE_SIZE=25000 # Characters

# Embedding settings
# EMBED_MESSAGES=true
# MAX_EMBEDDING_TOKENS=8192
# MAX_EMBEDDING_TOKENS_PER_REQUEST=300000

# LANGFUSE_HOST=
# LANGFUSE_PUBLIC_KEY=

COLLECT_METRICS_LOCAL=false
# LOCAL_METRICS_FILE=metrics.jsonl
# REASONING_TRACES_FILE=traces.jsonl  # Path to JSONL file for reasoning traces

# NAMESPACE="honcho"

# =============================================================================
# Database Settings (REQUIRED)
# =============================================================================
# Connection URI for PostgreSQL database with pgvector support
# Must use postgresql+psycopg prefix for SQLAlchemy compatibility
DB_CONNECTION_URI=postgresql+psycopg://postgres:postgres@localhost:5432/postgres

# Optional database settings
# DB_SCHEMA=public
# DB_POOL_CLASS=default
# DB_POOL_SIZE=10
# DB_MAX_OVERFLOW=20
# DB_POOL_TIMEOUT=30
# DB_POOL_RECYCLE=300
# DB_POOL_PRE_PING=true
# DB_POOL_USE_LIFO=true
# DB_SQL_DEBUG=false
# DB_TRACING=false

# =============================================================================
# Authentication Settings
# =============================================================================
# Whether to enable authentication (set to true for production)
AUTH_USE_AUTH=false

# JWT secret key (REQUIRED if AUTH_USE_AUTH=true)
# Generate with: python scripts/generate_jwt_secret.py
# AUTH_JWT_SECRET=your-secret-key-here

# =============================================================================
# LLM Provider (REQUIRED)
# =============================================================================
# Honcho uses LLMs for memory extraction, summarization, dialectic chat, and
# dream consolidation. The server will fail to start without a provider configured.
#
# Quick start: uncomment the two lines below, set your endpoint and API key,
# then uncomment the provider/model lines in each feature section below.
# Any OpenAI-compatible endpoint works (OpenRouter, Together, Fireworks, etc.).
# Models must support tool calling (function calling).
#
#LLM_OPENAI_COMPATIBLE_BASE_URL=https://openrouter.ai/api/v1
LLM_OPENAI_COMPATIBLE_API_KEY=none
#
# Provider options for each feature: custom, vllm, google, anthropic, openai, groq
# "custom" routes through the OpenAI-compatible endpoint above.
# Model name format depends on your provider (e.g., OpenRouter: vendor/model-name).
#
# ---- Alternative: vLLM self-hosted ------------------------------------------
# LLM_VLLM_BASE_URL=http://localhost:8000/v1
# LLM_VLLM_API_KEY=not-needed
#
# ---- Alternative: direct vendor keys (no endpoint needed) -------------------
# LLM_GEMINI_API_KEY=
# LLM_ANTHROPIC_API_KEY=
# LLM_OPENAI_API_KEY=
# LLM_GROQ_API_KEY=
#
# ---- General LLM settings ---------------------------------------------------
# Embedding provider — defaults to openai (requires LLM_OPENAI_API_KEY).
# Set to openrouter to route embeddings through your custom endpoint instead.
LLM_EMBEDDING_PROVIDER=openrouter
LLM_EMBEDDING_API_KEY=none
# LLM_DEFAULT_MAX_TOKENS=2500
# LLM_MAX_TOOL_OUTPUT_CHARS=10000
# LLM_MAX_MESSAGE_CONTENT_CHARS=2000

# =============================================================================
# Deriver (Background Worker)
# =============================================================================
DERIVER_ENABLED=true
DERIVER_PROVIDER=custom
#DERIVER_MODEL=your-model-here  # e.g. google/gemini-2.5-flash
# DERIVER_THINKING_BUDGET_TOKENS=1024  # gt=0 required; omit for non-thinking models
# DERIVER_WORKERS=1
# DERIVER_POLLING_SLEEP_INTERVAL_SECONDS=1.0
# DERIVER_STALE_SESSION_TIMEOUT_MINUTES=5
# DERIVER_QUEUE_ERROR_RETENTION_SECONDS=2592000
# DERIVER_TEMPERATURE=
# DERIVER_DEDUPLICATE=true
# DERIVER_MAX_OUTPUT_TOKENS=4096
# DERIVER_LOG_OBSERVATIONS=false
# DERIVER_MAX_INPUT_TOKENS=23000
# DERIVER_WORKING_REPRESENTATION_MAX_OBSERVATIONS=100
# DERIVER_REPRESENTATION_BATCH_MAX_TOKENS=1024
# DERIVER_FLUSH_ENABLED=false

# =============================================================================
# Peer Card
# =============================================================================
# PEER_CARD_ENABLED=true

# =============================================================================
# Dialectic
# =============================================================================
# DIALECTIC_MAX_OUTPUT_TOKENS=8192
# DIALECTIC_MAX_INPUT_TOKENS=100000
# DIALECTIC_HISTORY_TOKEN_LIMIT=8192
# DIALECTIC_SESSION_HISTORY_MAX_TOKENS=4096
#
# Per-level provider, model, and tuning:
DIALECTIC_LEVELS__minimal__PROVIDER=custom
#DIALECTIC_LEVELS__minimal__MODEL=your-model-here  # e.g. google/gemini-2.5-flash
DIALECTIC_LEVELS__minimal__THINKING_BUDGET_TOKENS=1
DIALECTIC_LEVELS__minimal__MAX_TOOL_ITERATIONS=1
DIALECTIC_LEVELS__minimal__MAX_OUTPUT_TOKENS=250
DIALECTIC_LEVELS__low__PROVIDER=custom
#DIALECTIC_LEVELS__low__MODEL=your-model-here
DIALECTIC_LEVELS__low__THINKING_BUDGET_TOKENS=1
DIALECTIC_LEVELS__low__MAX_TOOL_ITERATIONS=5
DIALECTIC_LEVELS__medium__PROVIDER=custom
#DIALECTIC_LEVELS__medium__MODEL=your-model-here
DIALECTIC_LEVELS__medium__THINKING_BUDGET_TOKENS=1
DIALECTIC_LEVELS__medium__MAX_TOOL_ITERATIONS=2
DIALECTIC_LEVELS__high__PROVIDER=custom
#DIALECTIC_LEVELS__high__MODEL=your-model-here
DIALECTIC_LEVELS__high__THINKING_BUDGET_TOKENS=1
DIALECTIC_LEVELS__high__MAX_TOOL_ITERATIONS=4
DIALECTIC_LEVELS__max__PROVIDER=custom
#DIALECTIC_LEVELS__max__MODEL=your-model-here
DIALECTIC_LEVELS__max__THINKING_BUDGET_TOKENS=1
DIALECTIC_LEVELS__max__MAX_TOOL_ITERATIONS=10

# =============================================================================
# Summary
# =============================================================================
# SUMMARY_ENABLED=true
SUMMARY_PROVIDER=custom
#SUMMARY_MODEL=your-model-here  # e.g. google/gemini-2.5-flash
SUMMARY_THINKING_BUDGET_TOKENS=512  # gt=0 required; omit for non-thinking models
SUMMARY_MESSAGES_PER_SHORT_SUMMARY=20
SUMMARY_MESSAGES_PER_LONG_SUMMARY=60
SUMMARY_MAX_TOKENS_SHORT=1000
SUMMARY_MAX_TOKENS_LONG=4000

# =============================================================================
# Dream
# =============================================================================
# DREAM_ENABLED=true
DREAM_PROVIDER=custom
#DREAM_MODEL=your-model-here  # e.g. google/gemini-2.5-flash
#DREAM_DEDUCTION_MODEL=your-model-here
#DREAM_INDUCTION_MODEL=your-model-here
DREAM_THINKING_BUDGET_TOKENS=8192  # gt=0 required; omit for non-thinking models
# DREAM_DOCUMENT_THRESHOLD=50
# DREAM_IDLE_TIMEOUT_MINUTES=60
# DREAM_MIN_HOURS_BETWEEN_DREAMS=8
# DREAM_ENABLED_TYPES=["omni"]
# DREAM_MAX_OUTPUT_TOKENS=16384
# DREAM_MAX_TOOL_ITERATIONS=20
# DREAM_HISTORY_TOKEN_LIMIT=16384
#
# Surprisal sampling (advanced):
# DREAM_SURPRISAL__ENABLED=false
# DREAM_SURPRISAL__TREE_TYPE=kdtree
# DREAM_SURPRISAL__TREE_K=5
# DREAM_SURPRISAL__SAMPLING_STRATEGY=recent
# DREAM_SURPRISAL__SAMPLE_SIZE=200
# DREAM_SURPRISAL__TOP_PERCENT_SURPRISAL=0.10
# DREAM_SURPRISAL__MIN_HIGH_SURPRISAL_FOR_REPLACE=10
# DREAM_SURPRISAL__INCLUDE_LEVELS=["explicit","deductive"]

# =============================================================================
# Webhook Settings
# =============================================================================
# WEBHOOK_SECRET=
# WEBHOOK_MAX_WORKSPACE_LIMIT=10

# =============================================================================
# Monitoring and Observability (Optional)
# =============================================================================
# Sentry error tracking
# SENTRY_ENABLED=false
# SENTRY_DSN=your-sentry-dsn-here
# SENTRY_RELEASE=your-release-semver
# SENTRY_ENVIRONMENT=development
# SENTRY_TRACES_SAMPLE_RATE=0.1
# SENTRY_PROFILES_SAMPLE_RATE=0.1

# =============================================================================
# Prometheus Metrics Settings (Pull-based metrics)
# =============================================================================
METRICS_ENABLED=true
# METRICS_NAMESPACE=honcho  # Inherits from NAMESPACE if not set

# =============================================================================
# CloudEvents Telemetry Settings (Analytics events)
# =============================================================================
# TELEMETRY_ENABLED=false
# TELEMETRY_ENDPOINT=https://telemetry.honcho.dev/v1/events
# TELEMETRY_HEADERS={"Authorization": "Bearer your-token"}  # JSON string for auth headers
# TELEMETRY_BATCH_SIZE=100
# TELEMETRY_FLUSH_INTERVAL_SECONDS=1.0
# TELEMETRY_FLUSH_THRESHOLD=50
# TELEMETRY_MAX_RETRIES=3
# TELEMETRY_MAX_BUFFER_SIZE=10000
# TELEMETRY_NAMESPACE=honcho  # Inherits from NAMESPACE if not set

# =============================================================================
# Cache
# =============================================================================
CACHE_ENABLED=true
# CACHE_URL="redis://localhost:6379/0?suppress=true"
# CACHE_NAMESPACE="honcho"  # Inherits from NAMESPACE if not set
# CACHE_DEFAULT_TTL_SECONDS=300
# CACHE_DEFAULT_LOCK_TTL_SECONDS=5

# =============================================================================
# Vector Store Settings
# =============================================================================
VECTOR_STORE_TYPE=pgvector  # pgvector, turbopuffer, lancedb
VECTOR_STORE_MIGRATED=true # true= migrated

# Global namespace prefix for all vector namespaces
# Namespaces follow the pattern: {NAMESPACE}.{type}.{hash}
# where hash is a base64url-encoded SHA-256 of the workspace/peer names
# - Documents: {NAMESPACE}.doc.{hash(workspace, observer, observed)}
# - Messages: {NAMESPACE}.msg.{hash(workspace)}
# VECTOR_STORE_NAMESPACE=honcho  # Inherits from NAMESPACE if not set

# Embedding dimensions (default: 1536 for OpenAI text-embedding-3-small)
VECTOR_STORE_DIMENSIONS=768

# Turbopuffer-specific settings (required if TYPE is "turbopuffer")
# VECTOR_STORE_TURBOPUFFER_API_KEY=your-turbopuffer-api-key
# VECTOR_STORE_TURBOPUFFER_REGION=gcp-us-east4

# LanceDB-specific settings (local embedded mode)
# VECTOR_STORE_LANCEDB_PATH=./lancedb_data

# Reconciliation interval for background sync (default: 5 minutes)
# VECTOR_STORE_RECONCILIATION_INTERVAL_SECONDS=300

LM Studio log shows requests/responses with Honcho : related to openai/text-embedding-3-small

2026-04-20 10:31:36  [INFO]
 Received request to embed multiple:  [
  "embedding test",
  "Sure thing! What would you like to embed?"
]
2026-04-20 10:31:36  [INFO]
 Returning embeddings (not shown in logs)
2026-04-20 10:31:37 [DEBUG]
 Received request: POST to /v1/embeddings with body  {
  "input": [
    "embedding test",
    "Sure thing! What would you like to embed?"
  ],
  "model": "openai/text-embedding-3-small",
  "encoding_format": "base64"
}
2026-04-20 10:31:37  [INFO]
 Received request to embed multiple:  [
  "embedding test",
  "Sure thing! What would you like to embed?"
]
2026-04-20 10:31:37  [INFO]
 Returning embeddings (not shown in logs)
2026-04-20 10:31:38 [DEBUG]
 Received request: POST to /v1/embeddings with body  {
  "input": [
    "embedding test",
    "Sure thing! What would you like to embed?"
  ],
  "model": "openai/text-embedding-3-small",
  "encoding_format": "base64"
}
2026-04-20 10:31:38  [INFO]
 Received request to embed multiple:  [
  "embedding test",
  "Sure thing! What would you like to embed?"
]
2026-04-20 10:31:38  [INFO]
 Returning embeddings (not shown in logs)

Honcho API log shows errors related to : expected 1536 dimensions, not 768

2026-04-20 10:16:58,451 - __main__ - INFO - Starting deriver queue processor
2026-04-20 10:16:58,452 - __main__ - INFO - Prometheus metrics server started on port 9090
2026-04-20 10:16:58,452 - __main__ - INFO - Running main loop
2026-04-20 10:16:58,456 - src.cache.client - INFO - Connected to cache at redis://redis:6379/0?suppress=true
2026-04-20 10:16:58,456 - src.reconciler.scheduler - INFO - ReconcilerScheduler started with 2 tasks: ['sync_vectors', 'cleanup_queue']
2026-04-20 10:21:59,466 - src.reconciler.scheduler - INFO - Enqueued reconciler task: sync_vectors
2026-04-20 10:22:01,450 - src.reconciler.sync_vectors - INFO - Vector reconciliation cycle completed (pgvector mode)
2026-04-20 10:27:00,460 - src.reconciler.scheduler - INFO - Enqueued reconciler task: sync_vectors
2026-04-20 10:27:02,387 - src.reconciler.sync_vectors - INFO - Vector reconciliation cycle completed (pgvector mode)
  File "/app/.venv/lib/python3.13/site-packages/starlette/middleware/base.py", line 168, in call_next
    raise app_exc from app_exc.__cause__ or app_exc.__context__
  File "/app/.venv/lib/python3.13/site-packages/starlette/middleware/base.py", line 144, in coro
    await self.app(scope, receive_or_disconnect, send_no_error)
  File "/app/.venv/lib/python3.13/site-packages/starlette/middleware/cors.py", line 87, in __call__
    await self.app(scope, receive, send)
  File "/app/.venv/lib/python3.13/site-packages/starlette/middleware/exceptions.py", line 63, in __call__
    await wrap_app_handling_exceptions(self.app, conn)(scope, receive, send)
  File "/app/.venv/lib/python3.13/site-packages/starlette/_exception_handler.py", line 53, in wrapped_app
    raise exc
  File "/app/.venv/lib/python3.13/site-packages/starlette/_exception_handler.py", line 42, in wrapped_app
    await app(scope, receive, sender)
  File "/app/.venv/lib/python3.13/site-packages/fastapi/middleware/asyncexitstack.py", line 18, in __call__
    await self.app(scope, receive, send)
  File "/app/.venv/lib/python3.13/site-packages/starlette/routing.py", line 716, in __call__
    await self.middleware_stack(scope, receive, send)
  File "/app/.venv/lib/python3.13/site-packages/starlette/routing.py", line 736, in app
    await route.handle(scope, receive, send)
  File "/app/.venv/lib/python3.13/site-packages/starlette/routing.py", line 290, in handle
    await self.app(scope, receive, send)
  File "/app/.venv/lib/python3.13/site-packages/fastapi/routing.py", line 119, in app
    await wrap_app_handling_exceptions(app, request)(scope, receive, send)
  File "/app/.venv/lib/python3.13/site-packages/starlette/_exception_handler.py", line 53, in wrapped_app
    raise exc
  File "/app/.venv/lib/python3.13/site-packages/starlette/_exception_handler.py", line 42, in wrapped_app
    await app(scope, receive, sender)
  File "/app/.venv/lib/python3.13/site-packages/fastapi/routing.py", line 105, in app
    response = await f(request)
               ^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.13/site-packages/fastapi/routing.py", line 431, in app
    raw_response = await run_endpoint_function(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    ...<3 lines>...
    )
    ^
  File "/app/.venv/lib/python3.13/site-packages/fastapi/routing.py", line 313, in run_endpoint_function
    return await dependant.call(**values)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/src/routers/messages.py", line 115, in create_messages_for_session
    "message_id": message.id,
                  ^^^^^^^^^^
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/attributes.py", line 569, in __get__
    return self.impl.get(state, dict_)  # type: ignore[no-any-return]
           ~~~~~~~~~~~~~^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/attributes.py", line 1096, in get
    value = self._fire_loader_callables(state, key, passive)
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/attributes.py", line 1126, in _fire_loader_callables
    return state._load_expired(state, passive)
           ~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/state.py", line 828, in _load_expired
    self.manager.expired_attribute_loader(self, toload, passive)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/loading.py", line 1674, in load_scalar_attributes
    result = load_on_ident(
        session,
    ...<4 lines>...
        no_autoflush=no_autoflush,
    )
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/loading.py", line 510, in load_on_ident
    return load_on_pk_identity(
        session,
    ...<11 lines>...
        is_user_refresh=is_user_refresh,
    )
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/loading.py", line 695, in load_on_pk_identity
    session.execute(
    ~~~~~~~~~~~~~~~^
        q,
        ^^
    ...<2 lines>...
        bind_arguments=bind_arguments,
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/session.py", line 2351, in execute
    return self._execute_internal(
           ~~~~~~~~~~~~~~~~~~~~~~^
        statement,
        ^^^^^^^^^^
    ...<4 lines>...
        _add_event=_add_event,
        ^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/session.py", line 2239, in _execute_internal
    conn = self._connection_for_bind(bind)
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/session.py", line 2108, in _connection_for_bind
    return trans._connection_for_bind(engine, execution_options)
           ~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<string>", line 2, in _connection_for_bind
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/state_changes.py", line 101, in _go
    self._raise_for_prerequisite_state(fn.__name__, current_state)
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/.venv/lib/python3.13/site-packages/sqlalchemy/orm/session.py", line 971, in _raise_for_prerequisite_state
    raise sa_exc.PendingRollbackError(
    ...<6 lines>...
    )
sqlalchemy.exc.PendingRollbackError: This Session's transaction has been rolled back due to a previous exception during flush. To begin a new transaction with this Session, first issue Session.rollback(). Original exception was: (builtins.ValueError) expected 1536 dimensions, not 768
[SQL: INSERT INTO public.message_embeddings (content, embedding, message_id, workspace_name, session_name, peer_name, sync_state, last_sync_at, sync_attempts) SELECT p0::VARCHAR, p1::VECTOR(1536), p2::VARCHAR, p3::VARCHAR, p4::VARCHAR, p5::VARCHAR, p6::VARCHAR, p7::TIMESTAMP WITH TIME ZONE, p8::INTEGER FROM (VALUES (%(content)s::VARCHAR, %(embedding)s, %(message_id)s::VARCHAR, %(workspace_name)s::VARCHAR, %(session_name)s::VARCHAR, %(peer_name)s::VARCHAR, %(sync_state)s::VARCHAR, %(last_sync_at)s::TIMESTAMP WITH TIME ZONE, %(sync_attempts)s::INTEGER)) AS imp_sen(p0, p1, p2, p3, p4, p5, p6, p7, p8, sen_counter) ORDER BY sen_counter RETURNING public.message_embeddings.id, public.message_embeddings.created_at, public.message_embeddings.id AS id__1]
[parameters: [{'message_id': 'PA8aY5ZeA-QEXwUvZxm0t', 'sync_state': 'pending', 'peer_name': 'owner', 'content': 'embedding test', 'workspace_name': 'main', 'session ... (16907 characters truncated) ... 544067144394, 0.007806772831827402, -0.0038657912518829107, -0.05361740663647652, -0.009963896125555038, 0.0022806485649198294], 'last_sync_at': None}, {'message_id': '0DOenfRYdieb0L2aPl-fo', 'sync_state': 'pending', 'peer_name': 'agent-main', 'content': 'Sure thing! What would you like to embed?', 'w ... (16963 characters truncated) ... 470942467451096, 0.04498995468020439, -0.03708455711603165, 0.0024789117742329836, 0.004620757419615984, -0.051529645919799805], 'last_sync_at': None}]] (Background on this error at: https://sqlalche.me/e/20/7s2a)
2026-04-20 10:32:01,461 - src.reconciler.scheduler - INFO - Enqueued reconciler task: sync_vectors
2026-04-20 10:32:03,464 - src.reconciler.sync_vectors - INFO - Vector reconciliation cycle completed (pgvector mode)

Image below :
openclaw (self hosted synapse/element/matrix) chat,
i say "embedding test" , then openclaw use LM Studio with Honcho,
honcho sends text-embedding-3-small request to LM Studio,
LM Studio responds with an other embedding model loaded manually (text-embedding-nomic-embed-text-v1.5@q8_0),
honsho crashes : "... Original exception was: (builtins.ValueError) expected 1536 dimensions, not 768 ..."
(honcho does not respect the config : VECTOR_STORE_DIMENSIONS=768 + custom embedding model)

sqlalchemy.exc.PendingRollbackError: This Session's transaction has been rolled back due to a previous exception during flush. To begin a new transaction with this Session, first issue Session.rollback(). Original exception was: (builtins.ValueError) expected 1536 dimensions, not 768
[SQL: INSERT INTO public.message_embeddings (content, embedding, message_id, workspace_name, session_name, peer_name, sync_state, last_sync_at, sync_attempts) SELECT p0::VARCHAR, p1::VECTOR(1536), p2::VARCHAR, p3::VARCHAR, p4::VARCHAR, p5::VARCHAR, p6::VARCHAR, p7::TIMESTAMP WITH TIME ZONE, p8::INTEGER FROM (VALUES (%(content)s::VARCHAR, %(embedding)s, %(message_id)s::VARCHAR, %(workspace_name)s::VARCHAR, %(session_name)s::VARCHAR, %(peer_name)s::VARCHAR, %(sync_state)s::VARCHAR, %(last_sync_at)s::TIMESTAMP WITH TIME ZONE, %(sync_attempts)s::INTEGER)) AS imp_sen(p0, p1, p2, p3, p4, p5, p6, p7, p8, sen_counter) ORDER BY sen_counter RETURNING public.message_embeddings.id, public.message_embeddings.created_at, public.message_embeddings.id AS id__1]
[parameters: [{'message_id': 'PA8aY5ZeA-QEXwUvZxm0t', 'sync_state': 'pending', 'peer_name': 'owner', 'content': 'embedding test', 'workspace_name': 'main', 'session ... (16907 characters truncated) ... 544067144394, 0.007806772831827402, -0.0038657912518829107, -0.05361740663647652, -0.009963896125555038, 0.0022806485649198294], 'last_sync_at': None}, {'message_id': '0DOenfRYdieb0L2aPl-fo', 'sync_state': 'pending', 'peer_name': 'agent-main', 'content': 'Sure thing! What would you like to embed?', 'w ... (16963 characters truncated) ... 470942467451096, 0.04498995468020439, -0.03708455711603165, 0.0024789117742329836, 0.004620757419615984, -0.051529645919799805], 'last_sync_at': None}]] (Background on this error at: https://sqlalche.me/e/20/7s2a)

Also it seems there is multiple tickets related to the same embedding hardcoded problem open, and they are not adressed ?
Not sure if my problem is exactly the same, so i did separate ticket.

(also i see many repos managed by AI, problems not solved, developpers unable to help because wibe coding degrading their skills, many dark code, and in some cases AI directly responding to requests and closing tickets without correcting bugs, hope this will not happen to this very promising repo/project)

Best regards :)

[dbeecham]

LLM_OPENAI_COMPATIBLE_BASE_URL <- this is the base url Honcho uses for both the LLM and the embedding agent

[dbeecham]

To use a separate embedding model, you can add this to config.py

src/config.py --- Python
217 217 
218 218     # General LLM settings
219 219     DEFAULT_MAX_TOKENS: Annotated[int, Field(default=1000, gt=0, le=100_000)] = 2500
... 220     EMBEDDING_MODEL: str | None = None
... 221     EMBEDDING_BASE_URL: str | None = None

Then, in src/embedding_client.py, line 65, you can use base_url=settings.LLM.EMBEDDING_BASE_URL or settings.LLM.OPENAI_COMPATIBLE_BASE_URL

And right under it set self.model = settings.LLM.EMBEDDING_MODEL or "text-embedding-3-small"

[warlordattack]

would be very nice them to ship this as an update to docker container

[dbeecham]

Yes. It's a commonly requested feature it seems. #590 #578 #564 #554 #518

[VVoruganti]

Closed by #678

[globalassist]

I noticed the same pattern in my deriver logs as shown in the snippet here: the reconciler registers both sync_vectors and cleanup_queue at startup (ReconcilerScheduler started with 2 tasks: [sync_vectors, cleanup_queue]), but only sync_vectors ever gets enqueued in practice.

As a result, the queue accumulates pending representation items that are never consumed. I wrote up my findings in more detail on #494 (#494 (comment)) with log excerpts and a workaround.