Meta DM / comment handling: cut your Claude bill 60-80%.
A 272 KB safety device answers what it can answer, blocks what it shouldn't. Gateway / Threads / Messenger / WhatsApp / Instagram DM / Graph API ― 6 routes. MIT source, 30-day free trial.
Concrete example: 10,000 DMs/month from $1,200 → $340 (~71% cut). Details on the service page.
Integrations list
Public Facebook Login + Claude
▸ Stop wasted shots through your internal ChatGPT
Prompt Gateway
Log in with a Meta account → in-browser RLM produces a D / μ / R verdict → only R goes to the Claude API = structurally lower API cost. All processing in-browser, no server. Live pipeline visualization included.
Public Threads Graph API + RLM
▸ Stop a viral-risk post before it ships
Threads automated posting (µ-prefilter)
Draft → in-browser RLM emits D / µ / R → on µ, warn before posting; on D/R, post via the Threads API. Access token lives in localStorage only, never touches the javatel server. Dry-run (curl preview) supported.
Public Messenger Send API + RLM
▸ Handle late-night inbound with instant FAQ replies
Messenger safe bot (D/µ/R pre-filter)
Incoming messages classified by RLM into D (instant FAQ) / µ (suppressed reply) / R (Claude delegation) → bot replies automatically. Page Access Token + PSID enables real Send API calls; works as a simulator without them. Dry-run (curl preview) supported.
Public WhatsApp Cloud API + RLM
▸ Auto-reply within the 24h rule, regulation-aware
WhatsApp Business safe bot (D/µ/R pre-filter)
On the WhatsApp Cloud API, RLM classifies incoming messages into D (instant FAQ) / µ (suppressed) / R (Claude delegation) → auto-reply. Access Token + Phone Number ID + E.164 phone for real Send; the 24h window + template policy are explicit. Simulator-only mode also works.
Public Instagram Graph API + RLM
▸ Discard spam DMs without ever calling the AI
Instagram DM safe bot (D/µ/R pre-filter)
RLM classifies incoming Instagram Business account DMs into D (instant FAQ) / µ (spam / speculation / PII-request suppression) / R (Claude delegation). Access Token + IG User ID + IGSID enables real Send; DM-centric design with explicit Story-reply / Comment-reply scopes. Honors the 24h window + HUMAN_AGENT tag constraints.
Public Graph API generic + RLM post-stage
▸ Multi-app debug + ops from a single panel
Graph API generic client
Swiss-army-knife UI that hits any Meta Graph API (Facebook / Instagram / Pages / Threads / WhatsApp) from one panel. Post-stage RLM analysis of response text into D/µ/R, curl-equivalent display, Webhook verification helper, history. For cross-app debugging and operations.
NEXT THEME · 2026-05-30 Meta = LLM-neutral Cross-vendor mix-and-match
★ What multi-agent extension changes ✅ Meta 6/6 B mode shipped
Meta integration is LLM-neutral ― any of Gemini / Claude / OpenAI, and crucially you can mix vendors between cheap and premium.
That freedom doesn't exist on X (Grok-only) or Google (Gemini-only); only Meta. Here is how the 6 multi-agent patterns (detailed on the Platform Integrations Hub) map onto Meta's 6 routes:
What only Meta offers
- Cross-vendor escalation ― cheap = Gemini Flash with premium = Claude Opus and the like; any vendor combination is fair game. Since Meta has no LLM lock-in, each vendor's strengths can be assigned per role.
- Individuals / SMBs run cheap on Gemini Flash's free tier (15 RPM) at $0, billing only for the 15% that escalates to premium → individuals run at single-digit dollars per month, enterprises run Opus-grade with a 95% cut ― same codebase.
- Cross-validation (A) and voting (F) are most natural on Meta ― send the same query to 3 vendors in parallel for a vote / mismatch detection. "Single-vendor dependency risk" is engineered out.
Common pattern
All Meta integrations share the same structure:
| Authentication | Meta Login for Business (OAuth 2.0) / minimal Meta App scopes public_profile + email |
|---|---|
| RLM implementation | Mock JS (public) lets you verify the API surface; production swaps in the actual WASM (~272 KB) |
| Verdict | D = instant (no Claude/LLM call) / μ = suppressed / R = LLM delegated |
| Audit | Every query chained via SHA-256; exportWAL() emits JSON → archive to S3 etc. |
| License | Integration code = MIT (public, modify and use commercially) / actual WASM = JAVATEL commercial license |
| URL convention | /resource/<product>/<integration>/<feature>/ ― e.g. /resource/slimetree-rlm/meta/gateway/ |