Fimeg/letta-server
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix: sanitize control characters before sending to inference backends

Browse Source 
Fireworks (via Synthetic Direct) chokes on raw ASCII control chars
(0x00-0x1F) in JSON payloads with "Unterminated string" errors.
The existing sanitize_unicode_surrogates only handles U+D800-DFFF.
Now we also strip control chars (preserving tab/newline/CR) at all
4 request paths — sync, async, and both streaming variants.
This commit is contained in:
Ani Tunturi
2026-03-21 20:23:56 -04:00
parent 2b3cfb1797
commit 08d3c26732
2 changed files with 36 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
letta/llm_api/openai_client.py
View File

@@ -28,7 +28,7 @@ from letta.errors import (
LLMTimeoutError,
LLMUnprocessableEntityError,
)
from letta.helpers.json_helpers import sanitize_unicode_surrogates
from letta.helpers.json_helpers import sanitize_control_characters, sanitize_unicode_surrogates
from letta.llm_api.error_utils import is_context_window_overflow_message, is_insufficient_credits_message
from letta.llm_api.helpers import (
add_inner_thoughts_to_functions,
@@ -669,6 +669,7 @@ class OpenAIClient(LLMClientBase):
"""
# Sanitize Unicode surrogates to prevent encoding errors
request_data = sanitize_unicode_surrogates(request_data)
request_data = sanitize_control_characters(request_data)
client = OpenAI(**self._prepare_client_kwargs(llm_config))
# Route based on payload shape: Responses uses 'input', Chat Completions uses 'messages'
@@ -694,6 +695,7 @@ class OpenAIClient(LLMClientBase):
"""
# Sanitize Unicode surrogates to prevent encoding errors
request_data = sanitize_unicode_surrogates(request_data)
request_data = sanitize_control_characters(request_data)
kwargs = await self._prepare_client_kwargs_async(llm_config)
client = AsyncOpenAI(**kwargs)
@@ -913,6 +915,7 @@ class OpenAIClient(LLMClientBase):
"""
# Sanitize Unicode surrogates to prevent encoding errors
request_data = sanitize_unicode_surrogates(request_data)
request_data = sanitize_control_characters(request_data)
kwargs = await self._prepare_client_kwargs_async(llm_config)
client = AsyncOpenAI(**kwargs)
@@ -947,6 +950,7 @@ class OpenAIClient(LLMClientBase):
"""
# Sanitize Unicode surrogates to prevent encoding errors
request_data = sanitize_unicode_surrogates(request_data)
request_data = sanitize_control_characters(request_data)
kwargs = await self._prepare_client_kwargs_async(llm_config)
client = AsyncOpenAI(**kwargs)