Commands

Type these as normal messages. The proxy intercepts them before the LLM.

Command Reference

Type these as normal messages in any client connected through the proxy. Case-insensitive, position-independent. Commands work on both proxy (streaming/non-streaming) and REST paths.

Command	Purpose
`VCATTACH`	Attach cross-platform memory to the current session
`VCLABEL`	Label the current conversation for future retrieval
`VCSTATUS`	Show context window state, active tags, and compression stats
`VCRECALL`	Recall a specific topic or time range
`VCCOMPACT`	Force immediate compaction
`VCLIST`	List available stored contexts and sessions
`VCFORGET`	Remove specific topics or facts from memory

VCATTACH

Attach stored memory from another session or platform to the current conversation. This is the mechanism for cross-platform memory: a conversation on Claude Code can access memories from a Telegram bot session, and vice versa.

Use Cases

Cross-platform continuity: Start a conversation on Telegram, continue it on Claude Code. Both sessions share the same memory pool.
Multi-agent memory: Multiple agents using the same virtual-context store share a unified fact base and topic index.
Session recovery: Reconnect to a previous session’s context after a fresh start.

How It Works

The command triggers engine.attach_contexts()
The engine queries the store for all sessions with stored segments
Segments are loaded into the retrieval index, facts into the fact store
The TurnTagIndex is updated to include tags from attached sessions
Subsequent retrieval queries can now surface content from any attached session

Attachment is additive. It does not remove or modify existing session content.

VCLABEL

Label the current conversation with a human-readable name. Labels are used for identification in VCLIST output and the dashboard.

VCLABEL MyProjectName

The label is stored as session metadata and persists across restarts.

VCSTATUS

Display the current state of the context window: conversation ID, label, turn count, compaction watermark, context window fill level, stored segments and token counts, active tags.

VCRECALL

VCRECALL the database migration discussion
VCRECALL what we talked about last Tuesday

Triggers retrieval outside the normal request flow. The engine parses the recall target, runs retrieval (or temporal resolution for time-based queries), and surfaces matching segments.

VCCOMPACT

Force immediate compaction regardless of threshold levels. Triggers the same compaction pipeline that runs automatically, but executes immediately. Protected recent turns are still preserved.

VCLIST

List all available stored contexts, sessions, and their metadata: session IDs with labels, segment counts, tag clouds, total stored tokens, compression ratios.

VCFORGET

VCFORGET the old API key discussion
VCFORGET fact: user lives in NYC

Remove specific topics or facts from memory. The engine identifies matching segments or facts and removes them from the store. This is a permanent deletion.

Cross-Platform Memory

Build up deep context in Claude Code: architecture decisions, code patterns, debugging history. Type VCATTACH in a Telegram conversation and immediately get that full context.

Both clients share the same conversation identity. Messages from either platform enrich the same compacted knowledge base. This is shared memory across platforms and models.