Lora synthesizes topics, searches 17K+ Telegram-channel posts, and generates platform-tailored ideas & drafts. Built for any product that needs to think in public.
https://lora-api.erarta.ai · v1.1.0
Every request (except /health) needs the X-API-Key header.
curl https://lora-api.erarta.ai/health
# {"status":"ok","service":"lora-api","version":"1.1.0"}import httpx
BASE = "https://lora-api.erarta.ai"
H = {"X-API-Key": "<your-key>", "Content-Type": "application/json"}
r = httpx.post(f"{BASE}/research",
json={"topic": "vertical AI agents for retail"},
headers=H, timeout=60).json()
print(r["research"]["answer"]) # Nova Pro synthesis
for n in r["channel_neighbors"]: # real TG posts
print(f"{n['views']} views: {n['text'][:120]}...")Synthesize a topic via AWS Bedrock Nova Pro (fallback: Gemini 2.5 Flash) plus semantic search across 17K+ Telegram-channel messages.
{
"topic": "pgvector vs Pinecone in production",
"include_channels": true,
"limit_neighbors": 5
}Response: research.answer, research.model (nova-pro or gemini-2.5-flash), channel_neighbors[] with text, date, views.
Full-text search over the corpus. Faster & cheaper than /research — no LLM synthesis.
{
"query": "embedding cache",
"channel_handle": "@ai_newz", # optional filter
"limit": 10
}List all tracked channels (50+, categories: ai, ml, startup, tools).
curl "https://lora-api.erarta.ai/channels?category=ai" -H "X-API-Key: <key>"Generate platform-tailored post ideas.
{
"topic": "how we replaced Pinecone with pgvector",
"platform": "linkedin", # telegram | linkedin | instagram
"count": 3
}Generate a full draft with built-in platform rules (tone, length, hashtags).
{
"idea": "Pinecone → pgvector: 10× cheaper, same latency",
"platform": "linkedin",
"tone": "direct, dry, first-person practitioner", # optional
"voice_rules": "avoid em-dashes, prefer numbers" # optional
}Calling /posts/generate applies platform-tested rules from May 2026 algorithm research:
| Platform | Language | Length | Hashtags |
|---|---|---|---|
| telegram | RU only | 600–1200 | 0 |
| EN only | 1300–2000 | 0–2 (3+ = −70% reach) | |
| EN only | 800–1500 | 3–5 (Dec 2025 cap) |
LinkedIn template grounded in Richard van der Blom's May 2026 1.8M-post analysis: Save = 5–10× a like, comment 15+ words = 15×, Topic Authority Score, em-dash penalty, engagement-bait shadow-ban patterns.
| Endpoint | Rate limit |
|---|---|
/research | 10 req/min |
/ideas | 20 req/min |
/posts/generate | 20 req/min |
/channels* | 30 req/min |
| Code | Meaning |
|---|---|
401 | Invalid X-API-Key |
422 | Validation failed (check field constraints) |
429 | Rate limit exceeded |
500 | LLM or Supabase error |