chore: 0.9.0 release (#2728)

Author: sarahwooders

alembic/versions/c4eb5a907b38_add_file_controls_to_agent_state.py

@@ -0,0 +1,33 @@
+"""Add file controls to agent state
+
+Revision ID: c4eb5a907b38
+Revises: cce9a6174366
+Create Date: 2025-07-21 15:56:57.413000
+
+"""
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "c4eb5a907b38"
+down_revision: Union[str, None] = "cce9a6174366"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column("agents", sa.Column("max_files_open", sa.Integer(), nullable=True))
+    op.add_column("agents", sa.Column("per_file_view_window_char_limit", sa.Integer(), nullable=True))
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_column("agents", "per_file_view_window_char_limit")
+    op.drop_column("agents", "max_files_open")
+    # ### end Alembic commands ###

alembic/versions/ddecfe4902bc_add_prompts.py

@@ -0,0 +1,42 @@
+"""add prompts
+
+Revision ID: ddecfe4902bc
+Revises: c4eb5a907b38
+Create Date: 2025-07-21 15:58:13.357459
+
+"""
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "ddecfe4902bc"
+down_revision: Union[str, None] = "c4eb5a907b38"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table(
+        "prompts",
+        sa.Column("id", sa.String(), nullable=False),
+        sa.Column("prompt", sa.String(), nullable=False),
+        sa.Column("created_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True),
+        sa.Column("updated_at", sa.DateTime(timezone=True), server_default=sa.text("now()"), nullable=True),
+        sa.Column("is_deleted", sa.Boolean(), server_default=sa.text("FALSE"), nullable=False),
+        sa.Column("_created_by_id", sa.String(), nullable=True),
+        sa.Column("_last_updated_by_id", sa.String(), nullable=True),
+        sa.Column("project_id", sa.String(), nullable=True),
+        sa.PrimaryKeyConstraint("id"),
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table("prompts")
+    # ### end Alembic commands ###

letta/__init__.py

@@ -5,16 +5,18 @@
     __version__ = version("letta")
 except PackageNotFoundError:
     # Fallback for development installations
-    __version__ = "0.8.17"
+    __version__ = "0.9.0"
 
 if os.environ.get("LETTA_VERSION"):
     __version__ = os.environ["LETTA_VERSION"]
 
-
 # import clients
 from letta.client.client import RESTClient
 
-# imports for easier access
+# Import sqlite_functions early to ensure event handlers are registered
+from letta.orm import sqlite_functions
+
+# # imports for easier access
 from letta.schemas.agent import AgentState
 from letta.schemas.block import Block
 from letta.schemas.embedding_config import EmbeddingConfig

letta/agent.py

@@ -36,6 +36,7 @@
 from letta.llm_api.helpers import calculate_summarizer_cutoff, get_token_counts_for_messages, is_context_overflow_error
 from letta.llm_api.llm_api_tools import create
 from letta.llm_api.llm_client import LLMClient
+from letta.local_llm.constants import INNER_THOUGHTS_KWARG
 from letta.local_llm.utils import num_tokens_from_functions, num_tokens_from_messages
 from letta.log import get_logger
 from letta.memory import summarize_messages
@@ -548,8 +549,8 @@ def _handle_ai_response(
                 return messages, False, True  # force a heartbeat to allow agent to handle error
 
             # Check if inner thoughts is in the function call arguments (possible apparently if you are using Azure)
-            if "inner_thoughts" in function_args:
-                response_message.content = function_args.pop("inner_thoughts")
+            if INNER_THOUGHTS_KWARG in function_args:
+                response_message.content = function_args.pop(INNER_THOUGHTS_KWARG)
             # The content if then internal monologue, not chat
             if response_message.content and not nonnull_content:
                 self.interface.internal_monologue(response_message.content, msg_obj=messages[-1], chunk_index=chunk_index)

letta/agents/base_agent.py

@@ -122,7 +122,9 @@ def extract_dynamic_section(text):
             curr_dynamic_section = extract_dynamic_section(curr_system_message_text)
 
             # generate just the memory string with current state for comparison
-            curr_memory_str = agent_state.memory.compile(tool_usage_rules=tool_constraint_block, sources=agent_state.sources)
+            curr_memory_str = agent_state.memory.compile(
+                tool_usage_rules=tool_constraint_block, sources=agent_state.sources, max_files_open=agent_state.max_files_open
+            )
             new_dynamic_section = extract_dynamic_section(curr_memory_str)
 
             # compare just the dynamic sections (memory blocks, tool rules, directories)
@@ -149,6 +151,7 @@ def extract_dynamic_section(text):
                 archival_memory_size=num_archival_memories,
                 tool_rules_solver=tool_rules_solver,
                 sources=agent_state.sources,
+                max_files_open=agent_state.max_files_open,
             )
 
             diff = united_diff(curr_system_message_text, new_system_message_str)

letta/agents/voice_agent.py

@@ -153,6 +153,7 @@ async def step_stream(self, input_messages: List[MessageCreate], max_steps: int
             previous_message_count=self.num_messages,
             archival_memory_size=self.num_archival_memories,
             sources=agent_state.sources,
+            max_files_open=agent_state.max_files_open,
         )
         letta_message_db_queue = create_input_messages(
             input_messages=input_messages, agent_id=agent_state.id, timezone=agent_state.timezone, actor=self.actor

letta/constants.py

@@ -326,7 +326,7 @@ def FUNCTION_RETURN_VALUE_TRUNCATED(return_str, return_char: int, return_char_li
 CORE_MEMORY_PERSONA_CHAR_LIMIT: int = 5000
 CORE_MEMORY_HUMAN_CHAR_LIMIT: int = 5000
 CORE_MEMORY_BLOCK_CHAR_LIMIT: int = 5000
-CORE_MEMORY_SOURCE_CHAR_LIMIT: int = 50000
+
 # Function return limits
 FUNCTION_RETURN_CHAR_LIMIT = 6000  # ~300 words
 BASE_FUNCTION_RETURN_CHAR_LIMIT = 1000000  # very high (we rely on implementation)
@@ -361,7 +361,9 @@ def FUNCTION_RETURN_VALUE_TRUNCATED(return_str, return_char: int, return_char_li
 REDIS_RUN_ID_PREFIX = "agent:send_message:run_id"
 
 # TODO: This is temporary, eventually use token-based eviction
-MAX_FILES_OPEN = 5
+# File based controls
+DEFAULT_MAX_FILES_OPEN = 5
+DEFAULT_CORE_MEMORY_SOURCE_CHAR_LIMIT: int = 50000
 
 GET_PROVIDERS_TIMEOUT_SECONDS = 10
 

letta/functions/schema_generator.py

@@ -412,12 +412,13 @@ def generate_schema(function, name: Optional[str] = None, description: Optional[
     # Validate that the function has a Google Python style docstring
     try:
         validate_google_style_docstring(function)
-    except ValueError:
+    except ValueError as e:
         logger.warning(
             f"Function `{function.__name__}` in module `{function.__module__}` "
             f"{'(tool_id=' + tool_id + ') ' if tool_id else ''}"
             f"is not in Google style docstring format. "
             f"Docstring received:\n{repr(function.__doc__[:200]) if function.__doc__ else 'None'}"
+            f"\nError: {str(e)}"
         )
 
     # Get the signature of the function

letta/groups/dynamic_multi_agent.py

@@ -94,6 +94,7 @@ def step(
                 for name, agent_id in [(agents[agent_id].agent_state.name, agent_id) for agent_id in agent_id_options]:
                     if name.lower() in assistant_message.content.lower():
                         speaker_id = agent_id
+                assert speaker_id is not None, f"No names found in {assistant_message.content}"
 
                 # Sum usage
                 total_usage.prompt_tokens += usage_stats.prompt_tokens

letta/helpers/converters.py

@@ -1,4 +1,3 @@
-import base64
 from typing import Any, Dict, List, Optional, Union
 
 import numpy as np
@@ -43,7 +42,10 @@
     TerminalToolRule,
     ToolRule,
 )
+from letta.settings import DatabaseChoice, settings
 
+if settings.database_engine == DatabaseChoice.SQLITE:
+    import sqlite_vec
 # --------------------------
 # LLMConfig Serialization
 # --------------------------
@@ -272,22 +274,28 @@ def deserialize_message_content(data: Optional[List[Dict]]) -> List[MessageConte
 
 
 def serialize_vector(vector: Optional[Union[List[float], np.ndarray]]) -> Optional[bytes]:
-    """Convert a NumPy array or list into a base64-encoded byte string."""
+    """Convert a NumPy array or list into serialized format using sqlite-vec."""
     if vector is None:
         return None
     if isinstance(vector, list):
         vector = np.array(vector, dtype=np.float32)
+    else:
+        vector = vector.astype(np.float32)
 
-    return base64.b64encode(vector.tobytes())
+    return sqlite_vec.serialize_float32(vector.tolist())
 
 
 def deserialize_vector(data: Optional[bytes], dialect: Dialect) -> Optional[np.ndarray]:
-    """Convert a base64-encoded byte string back into a NumPy array."""
+    """Convert serialized data back into a NumPy array using sqlite-vec format."""
     if not data:
         return None
 
     if dialect.name == "sqlite":
-        data = base64.b64decode(data)
+        # Use sqlite-vec format
+        if len(data) % 4 == 0:  # Must be divisible by 4 for float32
+            return np.frombuffer(data, dtype=np.float32)
+        else:
+            raise ValueError(f"Invalid sqlite-vec binary data length: {len(data)}")
 
     return np.frombuffer(data, dtype=np.float32)