whisper-context

# Whisper Context (OpenClaw Skill)

Reduce OpenClaw API spend by shrinking the context you send to the model (delta compression + caching), while keeping long-term memory across sessions.

This skill provides a minimal Node-based helper (`whisper-context.mjs`) that OpenClaw agents can run to:

- Retrieve packed context for a user/session (`query_context`) with `compress: true` and `compression_strategy: "delta"`
- Persist the latest turn into long-term memory (`ingest_session`)
- Write/search memories (`memory_write`, `memory_search`)
- Run Oracle search/research (`oracle_search`)
- Fetch cost analytics (`get_cost_summary`)
- Inspect/warm cache (`cache_stats`, `cache_warm`)

## Install (ClawHub)

```bash
npx clawhub@latest install whisper-context
```

ClawHub installs the skill folder into your OpenClaw skills workspace (typically `~/.openclaw/workspace/skills/`).

## Setup

Set environment variables (where OpenClaw reads env for your agent):

```bash
WHISPER_CONTEXT_API_URL=https://context.usewhisper.dev
WHISPER_CONTEXT_API_KEY=YOUR_KEY
WHISPER_CONTEXT_PROJECT=openclaw-cost-optimization
```

Notes:

- `WHISPER_CONTEXT_API_URL` is optional (defaults to `https://context.usewhisper.dev`).
- `WHISPER_CONTEXT_PROJECT` can be a project slug/name.
- If the project does not exist yet, the helper will auto-create it in your org on first use.
- For best memory behavior, use stable `user_id` and `session_id` values (don’t hardcode them globally; derive them per user/session in your agent).

## Usage

All commands print JSON to stdout.

### Global flags

- `--project <slugOrName>`: override `WHISPER_CONTEXT_PROJECT`
- `--api_url <url>`: override `WHISPER_CONTEXT_API_URL`
- `--timeout_ms <n>`: request timeout (default: 30000)

### Tips for real agents (to actually slash spend)

- Always call `query_context` first and inject the returned `context` instead of re-sending your entire chat history.
- Keep `compress: true`, `compression_strategy: "delta"`, and `use_cache: true` (the...

# Whisper Context (OpenClaw Skill)

Slash OpenClaw context costs by avoiding full chat-history replay:

- Delta compression (send only what changed)
- Intelligent caching (repeat queries get cheaper)
- Long-term memory (store/recall facts, preferences, goals, events)

## Install (ClawHub)

```bash
npx clawhub@latest install whisper-context
```

## Setup

```bash
export WHISPER_CONTEXT_API_URL=https://context.usewhisper.dev
export WHISPER_CONTEXT_API_KEY=YOUR_KEY
export WHISPER_CONTEXT_PROJECT=openclaw-cost-optimization
```

## One-liners

Query packed context (use this before calling the model):

```bash
node whisper-context.mjs query_context --query "PASTE_USER_MESSAGE" --user_id "user-123" --session_id "sess-123"
```

Ingest memory after you reply:

```bash
echo '{ "user": "PASTE_USER_MESSAGE", "assistant": "PASTE_ASSISTANT_REPLY" }' | node whisper-context.mjs ingest_session --session_id "sess-123" --user_id "user-123" --turn_json -
```

Cost analytics:

```bash
node whisper-context.mjs get_cost_summary --start_date "2026-01-01T00:00:00.000Z"
```

Cache stats:

```bash
node whisper-context.mjs cache_stats
```

## What This Calls (API endpoints)

- Token savings happens on `POST /v1/context/query` using `compress: true`, `compression_strategy: "delta"`, and `use_cache: true`.
- Memory persistence uses `POST /v1/memory/ingest/session`.
- Cost reporting uses `GET /v1/cost/summary`.

## Screenshots

Add images here when you publish to ClawHub:

- `./assets/openclaw-skill.png`
- `./assets/dashboard-costs.png`