[FEATURE] Automated Chat Classification and Routing to Projects

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Feature Request: Automated Chat Classification and Routing to Projects

Problem Statement

As Claude usage grows, power users increasingly manage multiple active Projects (e.g. separate workstreams for product development, technical research, client work). Today, new chats always land in the general chat list regardless of their content — requiring manual triage to move them into the appropriate project.

While the bulk-move UI introduced in late 2025 is a welcome improvement, it still requires the user to periodically review and manually re-sort accumulated chats. For users running 3+ active projects, this creates a recurring housekeeping burden that interrupts flow.

Impact

• Organizational friction: Conversations that belong in a project sit in the general list, disconnected from the relevant project context and knowledge base
• Memory/context fragmentation: Claude maintains separate memory summaries per project. Mis-routed chats contribute to the wrong memory context, degrading the quality of project-aware responses over time
• Scales poorly: The problem compounds with usage — the more active a user is across multiple workstreams, the more triage work accumulates

Proposed Solution

Introduce an optional auto-classification feature that uses Claude to suggest or automatically route chats — both new and historical — into the appropriate project.

Part 1: Forward-looking routing (new chats)

• After a chat's first exchange, Claude infers which project (if any) the conversation most likely belongs to
• User is shown a non-intrusive suggestion: "This looks like it belongs in [Project Name] — move it?"
• User can accept, redirect to a different project, or dismiss
• Optional: a user-configurable setting to auto-route without prompting for users who trust the classification

Part 2: Backlog classification (historical chats)

Introduce a one-time (and repeatable) "Organise my chats" workflow accessible from the chat list:

1. User initiates the workflow from settings or the chat sidebar
2. Claude reviews unassigned chat titles and opening messages across the user's full history
3. Claude generates a proposed grouping: a list of chats mapped to suggested projects, with a confidence indicator per suggestion
4. User is presented with a review UI — a simple table showing: chat title | suggested project | confidence | accept/reassign/skip
5. User bulk-approves, adjusts individual assignments, and confirms
6. Chats are moved in one operation

This makes historical organisation tractable in a single focused session rather than requiring manual one-by-one triage.

Supporting capabilities (optional / future)

• Allow project-level descriptions to act as routing hints — the classifier uses the project name and description to improve match quality
• Classification confidence threshold — only surface suggestions above a configurable threshold
• API access to conversation metadata and project routing, enabling programmatic workflows for teams or power users

Alternative Solutions

For users who prefer deterministic control over AI inference, provide a project routing rules configuration:

• Each project can have a set of user-defined keywords, topics, or tags
• When a new chat is created, its title and first message are matched against all project rules
• On a match, the chat is either auto-moved or flagged for confirmation depending on user preference
• For historical chats, a "Run classifier now" action applies the current ruleset to all unassigned chats and surfaces a bulk-review UI (similar to Part 2 above)

This approach is fully transparent and predictable — users know exactly why a chat was routed where it was — and avoids any concern about AI misclassification for sensitive or ambiguous conversations. The tradeoff is that it requires upfront configuration and ongoing rule maintenance as projects evolve.

Both solutions could coexist: rules-based routing takes precedence where a rule matches, and AI-assisted classification handles the remainder.

Priority

Medium - Would be very helpful

Feature Category

Interactive mode (TUI)

Use Case Example

---

Use Cases

Use Case 1: The Multi-Workstream Professional

Persona: A product person managing 3–4 concurrent workstreams (e.g. product development, technical research, partner enablement, personal productivity). They use Claude daily across all areas and have accumulated hundreds of unassigned chats over 12+ months.

Scenario: After creating Projects for each workstream, they face a backlog of ~300 historical chats with no practical way to sort them. Manual triage using the bulk-move UI would require reading each chat title, making a judgement call, and repeating hundreds of times. They abandon the effort and their projects remain incomplete — meaning project memory summaries are missing large amounts of relevant context.

Outcome with this feature: They trigger "Organize my chats", review a pre-classified table in a single session, accept ~80% of suggestions in bulk, manually reassign the ambiguous 20%, and confirm. Their projects now reflect the actual shape of their work. Memory summaries become meaningfully accurate for the first time.

---

Use Case 2: The Consultant Onboarding to Projects

Persona: An independent consultant who uses Claude for client deliverables across 5–6 active engagements. They resisted creating Projects initially, preferring to work informally, but now want the persistent context and knowledge base features.

Scenario: Switching to a Projects-based workflow mid-engagement means their existing client-specific chats — research threads, draft reviews, strategy discussions — remain in the general list. Without a classification tool, their new projects start with empty context even though months of relevant conversation history exists. They would need to manually identify and move chats per client, a task they keep deferring.

Outcome with this feature: The backlog classifier identifies chats by client theme using project names and descriptions as routing hints. The consultant reviews a pre-sorted list per engagement, confirms assignments, and their projects immediately inherit the relevant history. Onboarding to Projects takes one session instead of weeks of deferred triage.

---

Use Case 3: The Team Lead Managing Shared Projects

Persona: A team lead on a Claude Team plan, managing shared projects used by multiple colleagues. New chats frequently get started outside of the correct project context — either because team members forget to initiate from within the project, or because a conversation starts casually and only later becomes project-relevant.

Scenario: Important discussion threads and research chats accumulate in team members' personal chat lists rather than the shared project, meaning the project knowledge base and memory summary are incomplete. The team lead has no visibility into which unassigned chats are relevant to shared projects, and colleagues don't prioritize the admin of moving them.

Outcome with this feature: Auto-classification surfaces a prompt after the first exchange — "This looks like it belongs in [Project Name] — move it?" — catching mis-routed chats at the point of creation rather than after the fact. For the team lead, the periodic backlog workflow lets them do a bulk sweep to catch anything that slipped through. Shared project context stays accurate with minimal overhead.

---

Use Case 4: The Power User with Sensitive Workstreams

Persona: A senior professional using Claude across both high-sensitivity workstreams (e.g. HR decisions, confidential strategy) and general everyday tasks. They want project organization but are cautious about AI inference making routing decisions for sensitive content.

Scenario: An AI-only classification approach creates anxiety — they don't want Claude inferring the topic of a confidential conversation and auto-assigning it somewhere visible. They need deterministic, transparent control over how chats are classified.

Outcome with this feature: The rules-based routing option lets them define explicit keyword patterns per project. Sensitive projects have narrow, precise rules; general projects have broader ones. They know exactly why each chat was routed, can audit the logic, and trust the system enough to enable it. The AI-assisted classifier is reserved for non-sensitive workstreams where inference is acceptable.

---

Use Case 5: The Enterprise Team Building Custom Workflows

Persona: An IT or operations team at an Enterprise organization deploying Claude at scale. They manage project structures centrally and want to automate chat organization as part of a broader knowledge management workflow.

Scenario: Without API access to conversation metadata or project assignment, they cannot build any programmatic routing logic. Every employee manages their own chat list independently, leading to inconsistent project hygiene across the organization and degraded memory context at the team level.

Outcome with this feature: API access to conversation metadata and project assignment endpoints allows the team to build a lightweight internal classification service — applying company-specific taxonomy to route chats into the correct team or client project. Organization-wide project hygiene improves without requiring each individual to manage their own triage.

Additional Context

Why This Fits Claude's Direction

Projects are positioned as persistent, context-aware workspaces. Auto-routing reinforces that model — chats end up where the relevant knowledge base and instructions already live, making every conversation more useful from the first message. It also strengthens the value of memory summaries per project by ensuring those summaries reflect the right conversations. Solving the backlog problem is particularly important for user adoption: without it, new Project users face a cold-start problem where their most valuable historical context remains stranded in the general chat list.

Workarounds Considered

No native workaround exists. Automation via browser scripting against the UI is technically possible but fragile and unsupported. There is no API surface for project or conversation management.

---

Submitted by a Pro (and Max) user managing multiple active product workstreams.