Current Datetime

Current Datetime injects the current date and time into each request before it is forwarded to the LLM. This gives the model reliable temporal context for queries that depend on knowing the current date or time — for example: "What happened today?", "How many business days are left this month?", or "What should I schedule for tomorrow?"

Without this module, LLMs have no knowledge of the current date beyond their training cutoff, often producing incorrect or vague temporal answers.

Stage

Injected Format

The module appends the following block to the last user message in the request:

{original user message}

[System Context: Timezone={timezone}, CurrentTime={datetime}]

If no user message exists in the request, the block is appended to the last message of any role.

Example

Original user message:

How many days until the end of the quarter?

After injection (timezone: Asia/Jakarta, format: ISO 8601):

How many days until the end of the quarter?

[System Context: Timezone=Asia/Jakarta, CurrentTime=2026-06-25T09:30:00+07:00]

Prompt Caching Optimization

Multi-Turn Agent Loop Stability (Tool Call ID Relay)

In autonomous agent workflows where the LLM executes function or tool calls across multiple round-trips:

1. Turn 1 (Fresh User Query): The initial user prompt is stamped with the current timestamp (e.g. 2026-09-06T14:00:00Z). When the LLM responds with tool calls, Infralo indexes this timestamp in the gateway cache under the unique tool_call_id / call_id.
2. Turn 2+ (Tool Execution Results): Client SDKs store raw user messages in their local memory buffers without server-injected suffixes. When the agent sends tool results back along with the raw history, Infralo extracts the active tool_call_ids, retrieves Turn 1's exact timestamp from the gateway cache, and re-attaches it to the user message.

This prevents the seconds/minutes from drifting between turns, ensuring the prefix sent to the LLM remains 100% byte-for-byte stable for prompt cache hits throughout the entire agent loop. This is natively supported across both the Chat Completions (/v1/chat/completions) and Responses API (/v1/responses) endpoints.

Why Scoped by Tool Call ID (Not Session ID)?

We deliberately scope the timestamp to the individual tool_call_id rather than the broader session_id:

• A single user chat session can last hours or days and encompass dozens of separate conversations.
• Tying timestamps to a session ID would cause cross-conversation cache bleed, freezing the date/time on questions asked hours later to the session's first turn.
• The tool call ID strictly bounds the lifespan of the timestamp to that exact execution loop. Once the agent completes and the user asks a new question, a fresh timestamp is resolved naturally.

Framework Compatibility & Tool ID Preservation

For Infralo to retrieve the Turn 1 timestamp during tool continuations, the client SDK must pass back the same tool call ID that the model returned.

Important Note for Agno Users

By default, Agno's OpenAIChat and OpenAILike classes enforce OpenAI's legacy tool ID regex (^call_) by reformatting non-OpenAI IDs (e.g. Anthropic toolu_...) into a zero-padded counter: call_00000000000000000000000000000000000.

This causes two issues:

1. Infralo Cache Miss: Infralo indexes toolu_..., but Agno queries call_000000....
2. Provider KV Cache Miss: Changing the tool use ID in the assistant message forces Claude to re-tokenize the assistant block instead of hitting Anthropic's prompt cache.

To preserve native provider IDs and maintain full prompt caching when using Agno with Infralo Gateway, disable Agno's ID reformatting at application startup:

from agno.utils.message import PROVIDER_TOOL_ID_CONFIG

# Allow Infralo Gateway to manage provider-native IDs without Agno mutating them
if "openai_chat" in PROVIDER_TOOL_ID_CONFIG:
    PROVIDER_TOOL_ID_CONFIG["openai_chat"]["prefix"] = None

Configuration

Timezone

The IANA timezone identifier used to resolve the current datetime. Defaults to UTC.

All standard IANA timezone identifiers are supported. If an unknown or invalid timezone is provided, the module automatically falls back to UTC.

Common options:

Datetime Format

The output format for the injected datetime string.

Example Use Cases

• Time-sensitive Q&A bots: Chatbots that answer questions like "What's today?" or "When does this promotion end?" need reliable temporal context.
• Scheduling assistants: Agents that help users book meetings, set reminders, or reason about deadlines.
• Daily digest generation: Pipelines that generate summaries anchored to the current date.
• Business logic validation: LLMs evaluating time-based rules (e.g., "Is this invoice still within the payment window?").