Requirements
- Target platform
- OpenClaw
- Install method
- Manual import
- Extraction
- Extract archive
- Prerequisites
- OpenClaw
- Primary doc
- SKILL.md
Tencent SkillHub ยท Productivity
Telnyx Rag
Semantic search and Q&A over workspace files using Telnyx Storage + AI embeddings. Index your memory, knowledge, and skills for natural language retrieval and AI-powered answers.
skill openclawclawhub Free
Semantic search and Q&A over workspace files using Telnyx Storage + AI embeddings. Index your memory, knowledge, and skills for natural language retrieval and AI-powered answers.
โฌ 0 downloads โ
0 stars Unverified but indexed
Install for OpenClaw
Quick setup
- Download the package from Yavira.
- Extract the archive and review SKILL.md first.
- Import or place the package into your OpenClaw setup.
Package facts
- Download mode
- Yavira redirect
- Package format
- ZIP package
- Source platform
- Tencent SkillHub
- What's included
- SKILL.md, ask.py, config.json, diagram.svg, embed.sh, search.py
Validation
- Use the Yavira download entry.
- Review SKILL.md after the package is downloaded.
- Confirm the extracted package contains the expected setup assets.
Install with your agent
Agent handoff
Hand the extracted package to your coding agent with a concrete install brief instead of figuring it out manually.
- Download the package from Yavira.
- Extract it into a folder your agent can access.
- Paste one of the prompts below and point your agent at the extracted folder.
New install
I downloaded a skill package from Yavira. Read SKILL.md from the extracted folder and install it by following the included instructions. Tell me what you changed and call out any manual steps you could not complete.
Upgrade existing
I downloaded an updated skill package from Yavira. Read SKILL.md from the extracted folder, compare it with my current installation, and upgrade it while preserving any custom configuration unless the package docs explicitly say otherwise. Summarize what changed and any follow-up checks I should run.
Trust & source
Release facts
- Source
- Tencent SkillHub
- Verification
- Indexed source record
- Version
- 1.0.1
Provenance
- Publisher
- teamtelnyx
- Source page
- View original listing
- Canonical URL
- Open canonical page
Documentation
Telnyx RAG Memory
Semantic search and RAG-powered Q&A over your OpenClaw workspace using Telnyx's native embedding, similarity search, and inference APIs.
Requirements
Your own Telnyx API Key โ each user/agent uses their own key Python 3.8+ โ stdlib only, no external dependencies Get your API key at portal.telnyx.com
Bucket Naming Convention
Use a consistent naming scheme so anyone can adopt this: openclaw-{agent-id} AgentBucketChief (main)openclaw-mainBob the Builderopenclaw-builderVoice agentopenclaw-voiceYour agentopenclaw-{your-id} Why? Predictable: anyone can find any agent's bucket Collision-free: scoped to agent, not person or team Discoverable: openclaw-* prefix groups all agent buckets in Telnyx Storage UI
Quick Start
cd ~/skills/telnyx-rag # Set YOUR Telnyx API key (each user/agent uses their own) echo 'TELNYX_API_KEY=KEY...' > .env # Run setup with validation ./setup.sh --check # Validate requirements first ./setup.sh # Full setup (uses bucket from config.json) # Search your memory ./search.py "What are my preferences?" # Ask questions (full RAG pipeline) ./ask.py "What is the porting process?"
What It Does
Indexes your workspace files (MEMORY.md, memory/*.md, knowledge/, skills/) Chunks large files intelligently (markdown by headers, JSON/Slack by threads) Embeds content automatically using Telnyx AI Searches using natural language queries with retry logic Answers questions using a full RAG pipeline (retrieve โ rerank โ generate) Prioritizes results from memory/ (your primary context) Incremental sync โ only uploads changed files Orphan cleanup โ removes deleted files from bucket
Option 1: Environment Variable
export TELNYX_API_KEY="KEY..." ./setup.sh
Option 2: .env File
echo 'TELNYX_API_KEY=KEY...' > .env ./setup.sh
Validation Mode
./setup.sh --check # Validate requirements without making changes
Custom Bucket Name
./setup.sh my-custom-bucket
Ask Questions (RAG Pipeline)
# Basic question answering ./ask.py "What is Telnyx's porting process?" # Show retrieved context alongside answer ./ask.py "How do I deploy?" --context # Use a different model ./ask.py "Explain voice setup" --model meta-llama/Meta-Llama-3.1-8B-Instruct # More/fewer context chunks ./ask.py "meeting decisions" --num 12 # JSON output for scripting ./ask.py "API usage limits" --json # Search a different bucket ./ask.py "project timeline" --bucket work-memory
Search Memory
# Basic search with improved error handling ./search.py "What are David's communication preferences?" # Search specific bucket ./search.py "meeting notes" --bucket my-other-bucket # More results with timeout control ./search.py "procedures" --num 10 --timeout 45 # JSON output (for scripts) ./search.py "procedures" --json
Sync Files (with Chunking)
# Incremental sync with auto-chunking ./sync.py # Override chunk size (tokens) ./sync.py --chunk-size 600 # Quiet mode for cron jobs ./sync.py --quiet # Remove orphaned files (including stale chunks) ./sync.py --prune # Sync + trigger embedding ./sync.py --embed # Check status ./sync.py --status # List indexed files (shows chunks too) ./sync.py --list
Watch Mode
# Watch for changes and auto-sync with chunking ./sync.py --watch
Trigger Embedding
# Trigger embedding for current bucket ./embed.sh # OR ./sync.py --embed # Check embedding status ./sync.py --embed-status <task_id> Why is this needed? Uploading files to Telnyx Storage doesn't automatically generate embeddings. The embedding process converts your files into searchable vectors. Without this step, search.py and ask.py won't return results.
Configuration
Edit config.json to customize behavior: { "bucket": "openclaw-memory", "region": "us-central-1", "workspace": ".", "patterns": [ "MEMORY.md", "memory/*.md", "knowledge/*.json", "skills/*/SKILL.md" ], "priority_prefixes": ["memory/", "MEMORY.md"], "default_num_docs": 5, "chunk_size": 800, "ask_model": "meta-llama/Meta-Llama-3.1-70B-Instruct", "ask_num_docs": 8, "retrieve_num_docs": 20 }
Config Fields
FieldDefaultDescriptionbucketopenclaw-{agent-id}Telnyx Storage bucket name (see naming convention)regionus-central-1Storage regionworkspace.Root directory to scan for filespatterns(see above)Glob patterns for files to indexpriority_prefixes["memory/", "MEMORY.md"]Sources to rank higher in resultsexclude["*.tmp", ...]Patterns to excludechunk_size800Target tokens per chunk (~4 chars/token)ask_modelMeta-Llama-3.1-70B-InstructLLM model for ask.pyask_num_docs8Final context chunks for LLMretrieve_num_docs20Initial retrieval count (before reranking)
How It Works
โโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ Your Workspace โ โ Telnyx Cloud โ โ โโโ memory/ โ โ โ โ โโโ knowledge/ โโโโ โ Storage: your-bucket/ โ โ โโโ skills/ โ โ โ โโโ file__chunk-001.md โ โโโโโโโโโโโโโโโโโโโ โ โ โโโ file__chunk-002.md โ โ โ โ โ Smart Chunking โโโโ โ โผ embed โ โโโ Markdown: split โ Telnyx AI Embeddings โ โ on ## headers โ โ โ โโโ JSON/Slack: split โ โผ โ โ by thread/time โ Similarity Search โ โโโ Metadata tags โ โ โ โโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโ โ ask.py Pipeline: โ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ โ 1. Retrieve top-20 chunks โโโโโโ โ โ 2. Rerank (TF-IDF + priority) โ โ 3. Deduplicate adjacent chunks โ โ 4. Build prompt with top-8 โ โ 5. Call Telnyx Inference LLM โ โ 6. Return answer + sources โ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
Smart Chunking
Large files are automatically split into semantic chunks before upload:
Markdown Files
Split on ## and ### headers first If a section is still too large, split by paragraph boundaries Each chunk gets a metadata header with source, chunk index, and title
JSON / Slack Exports
Messages grouped by token budget per chunk Extracts: channel name, date range, authors Metadata includes Slack-specific fields
Chunk Naming
Chunks use deterministic filenames: knowledge/meetings.md โ knowledge/meetings__chunk-001.md knowledge/meetings__chunk-002.md knowledge/meetings__chunk-003.md
Chunk Metadata
Each chunk includes a YAML-style header: --- source: knowledge/meetings.md chunk: 2/5 title: Q4 Planning Discussion --- (chunk content here) For Slack exports, additional fields: --- source: slack/general.json chunk: 3/12 title: general channel: general date_range: 2024-01-15 to 2024-01-16 authors: alice, bob, charlie ---
Chunk Lifecycle
When a source file changes, old chunks are deleted and new ones uploaded Chunk mappings tracked in .sync-state.json --prune cleans up orphaned chunks from deleted files
Reranking (ask.py)
The RAG pipeline uses a multi-signal reranking strategy: Semantic similarity โ Telnyx embedding distance (certainty score) Keyword overlap โ TF-IDF weighted term matching with the query Priority boost โ Chunks from priority_prefixes sources ranked higher Deduplication โ Adjacent chunks from the same source with >80% token overlap are merged Initial retrieval fetches retrieve_num_docs (default 20), reranking selects the best ask_num_docs (default 8) for the LLM prompt.
Smart Chunking
Semantic splitting: Headers for markdown, threads for Slack JSON Metadata headers: Source, chunk index, title in every chunk Configurable size: --chunk-size flag or chunk_size in config Deterministic names: Reproducible chunk filenames
RAG Q&A Pipeline (ask.py)
End-to-end: Query โ retrieve โ rerank โ generate โ answer Telnyx Inference: Uses Telnyx LLM API for generation Source references: Every answer includes source file citations Context mode: --context shows retrieved chunks JSON output: --json for structured responses
Reranking
Multi-signal scoring: Combines embedding similarity + keyword overlap + priority Deduplication: Removes near-identical adjacent chunks Configurable: Retrieve 20, use best 8 (tunable)
Incremental Sync (v1)
File hashing: Tracks SHA-256 hashes in .sync-state.json Skip unchanged: Only uploads modified files Progress tracking: Shows progress bars for large syncs
Smart Cleanup
--prune: Removes files from bucket that were deleted locally Chunk-aware: Cleans up orphaned chunks too State tracking: Maintains sync history and chunk mappings
Improved Reliability
Retry logic: 3 attempts with exponential backoff Better errors: Parses Telnyx API error responses Timeout control: Configurable request timeouts Quiet mode: --quiet flag for cron jobs
OpenClaw Integration
Add to your TOOLS.md: ## Semantic Memory & Q&A Ask questions about your workspace: \`\`\`bash cd ~/skills/telnyx-rag && ./ask.py "your question" \`\`\` Search memory semantically: \`\`\`bash cd ~/skills/telnyx-rag && ./search.py "your query" \`\`\`
Automated Sync
Add to your heartbeat or cron: # Quiet sync with orphan cleanup cd ~/skills/telnyx-rag && ./sync.py --quiet --prune # Sync with embedding cd ~/skills/telnyx-rag && ./sync.py --quiet --embed
Setup Issues
"Python version too old" Requires Python 3.8+ Check: python3 --version "API key test failed" Verify key: echo $TELNYX_API_KEY Get new key at portal.telnyx.com
Sync Issues
"Bucket not found" ./sync.py --create-bucket "No results found" Wait 1-2 minutes after sync (embeddings take time) Check files uploaded: ./sync.py --list Trigger embedding: ./sync.py --embed "Files not syncing" Check .sync-state.json for corruption Force re-sync: rm .sync-state.json && ./sync.py
Ask Issues
"LLM generation failed" Check API key has inference permissions Try a different model: ./ask.py "query" --model meta-llama/Meta-Llama-3.1-8B-Instruct "No relevant documents found" Ensure files are synced and embedded Try broader query terms
From Python
from ask import ask from search import search_memory # Ask a question (full RAG pipeline) answer = ask("What is the deployment process?") print(answer) # With options answer = ask( "project timeline", num_final=5, model="meta-llama/Meta-Llama-3.1-8B-Instruct", show_context=True, output_json=True, ) print(answer) # Basic search results = search_memory("What do I know about X?", num_docs=5) print(results)
From Bash
# Ask and capture answer answer=$(./ask.py "What are the API limits?" --json) # Search and capture JSON results=$(./search.py "query" --json)
Performance Tips
Tune chunk_size โ Smaller chunks (400-600) for precise retrieval, larger (800-1200) for more context Use --quiet for cron jobs to reduce output Enable --prune periodically to clean up deleted files Watch mode is great for development: ./sync.py --watch Batch embedding by syncing first, then embedding: ./sync.py && ./sync.py --embed
Credits
Built for OpenClaw using Telnyx Storage and AI APIs.
Category context
Workflow acceleration for inboxes, docs, calendars, planning, and execution loops.
Source: Tencent SkillHub
Largest current source with strong distribution and engagement signals.
Package contents
Included in package
3 Scripts1 Docs1 Config1 Files
- SKILL.md
- ask.py
- embed.sh
- search.py
- config.json
- diagram.svg