Memory tagging in Chronologue allows users and agents to organize, retrieve, and share meaningful subsets of memory traces. Tags act as lightweight semantic labels—designed to give users control, agents precision, and teams a shared language for context.

This document walks through the design motivations for tagging, the tension between structured control and intelligent retrieval, and the practical advantages of using tags for planning, reflection, and agent collaboration.

​
1. Why Tagging Matters

While language models can infer patterns from raw memory traces, Chronologue supports tagging for two core reasons:

​
a. Control Interface

Tags let users shape the context used in agent tasks. For example:

  • Summarize my goals from #q2_review
  • Schedule around #deep_work blocks
  • Reflect only on #travel memories this weekend

This is more than filtering—it’s a way to think with memory, not just about it.

​
b. Communication Interface

Tags create a common vocabulary between users, agents, and collaborators. This enables:

  • Sharing a memory cluster (#team_sync) with a new agent
  • Broadcasting #customer_feedback traces to a product team
  • Subscribing an agent to #urgent_tasks for autonomous triage

​
2. Schema Overview

Every memory trace in Chronologue can include an array of tags:

tags: [“deep_work”, “2025Q2”, “team_sync”, “urgent”]

These tags are:

  • Plain strings
  • User- or agent-defined
  • Stored alongside trace metadata

​
3. Use Cases

​
a. Personal Context Scoping

User prompt:

“Can you generate a personal retrospective based on my #weekly_reflection and #fitness logs from April?”

System behavior:

  • Fetches all memory traces with tags containing both weekly_reflection and fitness
  • Limits by timestamp from 2025-04-01 to 2025-04-30
  • Passes to LLM as structured context block

​
b. Multi-Agent Collaboration

  • A user tags a batch of customer support transcripts with #feature_request
  • Shares this tag cluster with a product roadmap agent and an investor report agent
  • Each agent retrieves the same core memory, but interprets it for different planning tasks

​
c. Cross-User Shared Memory

In a team workspace:

  • A lead tags a set of traces with #onboarding_guides
  • New team members can access this tag’s contents for reference
  • An onboarding agent uses only #onboarding_guides to generate customized start plans

​
4. Prompt-Aware Tag Filters

Chronologue supports #hashtags in chat prompts:

Prompt:

“Summarize everything in #deep_work and #2025Q1 related to travel planning.”

System:

  • Parses #deep_work and #2025Q1 as tag filters
  • Retrieves intersected trace cluster
  • Conditionally scopes memory for planning or reflection agents

This pattern allows users to dynamically reframe memory context with a single prompt.

​
5. Design Tension: Intelligence vs. Explicit Control

​
Argument Against Tagging

  • Modern LLMs can infer themes without tags
  • Embeddings can cluster memories without manual labeling
  • Tags require upkeep and structure

​
Why Chronologue Still Embraces Tagging

  • Tags allow repeatable, editable, and inspectable context control
  • They enable human-to-agent communication of memory intent
  • Tags support parallel views of the same memory (e.g., #urgent, #revisit, #weekly)

Tagging is not about redundancy—it’s about empowering users to decide what matters and when.

​
6. Best Practices

  • Use simple, memorable tags: #weekly, #team_sync, #urgent
  • Combine temporal and semantic tags: #may, #q2_review, #travel
  • Let agents suggest tags during or after execution (e.g., “Would you like to tag this #deep_work?”)
  • Regularly audit and summarize high-frequency tag clusters

​
7. Future Directions

  • Tag versioning and namespaces for shared workspaces
  • Auto-generated tags based on NLP themes and calendar context
  • Agent-driven tag propagation and conditional triggers (e.g., if tagged #urgent, notify)
  • Visual tag-based context composer for multi-agent systems