feat(copilot): E2B cloud sandbox — unified file tools, persistent execution, output truncation

autogpt_platform/.gitignore

@@ -1,2 +1,3 @@
 *.ignore.*
-*.ign.*
+*.ign.*
+.application.logs

autogpt_platform/backend/Dockerfile

@@ -95,7 +95,7 @@ ENV DEBIAN_FRONTEND=noninteractive
 
 # Install Python, FFmpeg, ImageMagick, and CLI tools for agent use.
 # bubblewrap provides OS-level sandbox (whitelist-only FS + no network)
-# for the bash_exec MCP tool.
+# for the bash_exec MCP tool (fallback when E2B is not configured).
 # Using --no-install-recommends saves ~650MB by skipping unnecessary deps like llvm, mesa, etc.
 RUN apt-get update && apt-get install -y --no-install-recommends \
     python3.13 \

autogpt_platform/backend/backend/copilot/config.py

@@ -93,12 +93,49 @@ class ChatConfig(BaseSettings):
         "history compression. Falls back to compression when unavailable.",
     )
 
+    # E2B Sandbox Configuration
+    use_e2b_sandbox: bool = Field(
+        default=True,
+        description="Use E2B cloud sandboxes for persistent bash/python execution. "
+        "When enabled, bash_exec routes commands to E2B and SDK file tools "
+        "operate directly on the sandbox via E2B's filesystem API.",
+    )
+    e2b_api_key: str | None = Field(
+        default=None,
+        description="E2B API key. Falls back to E2B_API_KEY environment variable.",
+    )
+    e2b_sandbox_template: str = Field(
+        default="base",
+        description="E2B sandbox template to use for copilot sessions.",
+    )
+    e2b_sandbox_timeout: int = Field(
+        default=43200,  # 12 hours — same as session_ttl
+        description="E2B sandbox keepalive timeout in seconds.",
+    )
+
     # Extended thinking configuration for Claude models
     thinking_enabled: bool = Field(
         default=True,
         description="Enable adaptive thinking for Claude models via OpenRouter",
     )
 
+    @field_validator("use_e2b_sandbox", mode="before")
+    @classmethod
+    def get_use_e2b_sandbox(cls, v):
+        """Get use_e2b_sandbox from environment if not provided."""
+        env_val = os.getenv("CHAT_USE_E2B_SANDBOX", "").lower()
+        if env_val:
+            return env_val in ("true", "1", "yes", "on")
+        return True if v is None else v
+
+    @field_validator("e2b_api_key", mode="before")
+    @classmethod
+    def get_e2b_api_key(cls, v):
+        """Get E2B API key from environment if not provided."""
+        if v is None:
+            v = os.getenv("CHAT_E2B_API_KEY") or os.getenv("E2B_API_KEY")
+        return v
+
     @field_validator("api_key", mode="before")
     @classmethod
     def get_api_key(cls, v):

autogpt_platform/backend/backend/copilot/db.py

@@ -16,7 +16,7 @@
 )
 
 from backend.data import db
-from backend.util.json import SafeJson
+from backend.util.json import SafeJson, sanitize_string
 
 from .model import ChatMessage, ChatSession, ChatSessionInfo
 
@@ -101,15 +101,16 @@ async def add_chat_message(
         "sequence": sequence,
     }
 
-    # Add optional string fields
+    # Add optional string fields — sanitize to strip PostgreSQL-incompatible
+    # control characters (null bytes etc.) that may appear in tool outputs.
     if content is not None:
-        data["content"] = content
+        data["content"] = sanitize_string(content)
     if name is not None:
         data["name"] = name
     if tool_call_id is not None:
         data["toolCallId"] = tool_call_id
     if refusal is not None:
-        data["refusal"] = refusal
+        data["refusal"] = sanitize_string(refusal)
 
     # Add optional JSON fields only when they have values
     if tool_calls is not None:
@@ -170,15 +171,16 @@ async def add_chat_messages_batch(
                         "createdAt": now,
                     }
 
-                    # Add optional string fields
+                    # Add optional string fields — sanitize to strip
+                    # PostgreSQL-incompatible control characters.
                     if msg.get("content") is not None:
-                        data["content"] = msg["content"]
+                        data["content"] = sanitize_string(msg["content"])
                     if msg.get("name") is not None:
                         data["name"] = msg["name"]
                     if msg.get("tool_call_id") is not None:
                         data["toolCallId"] = msg["tool_call_id"]
                     if msg.get("refusal") is not None:
-                        data["refusal"] = msg["refusal"]
+                        data["refusal"] = sanitize_string(msg["refusal"])
 
                     # Add optional JSON fields only when they have values
                     if msg.get("tool_calls") is not None:
@@ -312,7 +314,7 @@ async def update_tool_message_content(
                 "toolCallId": tool_call_id,
             },
             data={
-                "content": new_content,
+                "content": sanitize_string(new_content),
             },
         )
         if result == 0:

autogpt_platform/backend/backend/copilot/response_model.py

@@ -13,6 +13,7 @@
 from pydantic import BaseModel, Field
 
 from backend.util.json import dumps as json_dumps
+from backend.util.truncate import truncate
 
 logger = logging.getLogger(__name__)
 
@@ -150,6 +151,9 @@ class StreamToolInputAvailable(StreamBaseResponse):
     )
 
 
+_MAX_TOOL_OUTPUT_SIZE = 100_000  # ~100 KB; truncate to avoid bloating SSE/DB
+
+
 class StreamToolOutputAvailable(StreamBaseResponse):
     """Tool execution result."""
 
@@ -164,6 +168,10 @@ class StreamToolOutputAvailable(StreamBaseResponse):
         default=True, description="Whether the tool execution succeeded"
     )
 
+    def model_post_init(self, __context: Any) -> None:
+        """Truncate oversized outputs after construction."""
+        self.output = truncate(self.output, _MAX_TOOL_OUTPUT_SIZE)
+
     def to_sse(self) -> str:
         """Convert to SSE format, excluding non-spec fields."""
         data = {