Agent Infrastructure

The Agent Safety
Layer

Memory + Cost Control + Gateway in one integrated suite. Agents that remember, stay in budget, and execute reliably — with any framework.

Free · no credit card · instant key
Works with LangChain CrewAI Claude SDK OpenAI Agents Any HTTP client
AES-256 Encrypted
No Data Sharing
REST API — any stack
4 Core Endpoints
<100ms P95 Latency
Free No Credit Card
SDK npm install sendryx-js

Up in 3 steps

From zero to persistent agent memory in under 5 minutes.

01

Get Your API Key

Enter your email above. An API key hits your inbox instantly — free forever, no card required.

X-API-Key: sk_live_…
02

Store Agent Memory

POST any context — facts, user prefs, past decisions — and Sendryx indexes it semantically.

POST /api/memory/store
03

Search Semantically

Query in plain language. Get back the most relevant memories, ranked by meaning — not keyword.

POST /api/memory/search
SDK shortcut npm install sendryx-js Full API docs →

Three services. One safety layer.

No competitor owns all three. Sendryx is the only integrated suite that covers every risk point in a production agent stack.

🔀

Gateway

A transparent proxy between your agents and every external API. Rate limiting, secret scrubbing, and full request logging — on every call, by default.

Explore Gateway →
⚙️

Cost Control

Set spend caps per namespace. Automatic circuit breakers stop runaway agents at the limit — not after the bill arrives. Webhook alerts at 80% and 100%.

Explore Cost Control →
🧠

Memory API Live

Persistent, searchable memory for any agent. Store facts, retrieve context, forget on command. Cross-session, cross-agent, framework agnostic.

Explore Memory API →

Where the competition falls short

Every agent infrastructure tool scores an F on memory. Sendryx is the only one that solves all three layers.

Capability Sendryx Helicone Portkey LangSmith
AI Gateway / Proxy A B A F
Cost Control & Budget Caps A C B F
Persistent Agent Memory A F F F
Framework Agnostic A B B C
Secret Scrubbing A F F F
Observability & Logging A A B A
The only suite that covers all three layers.
Get your API key in seconds. Free tier, no credit card.

How we're different

vs Portkey

Portkey is the gateway. Sendryx is the safety layer on top.

vs Helicone

Helicone observes what happened. Sendryx prevents what shouldn't happen.

vs LangSmith

LangSmith optimizes LangGraph. Sendryx optimizes any agent.

Three failure modes. One fix.

Every production agent breaks in the same three ways. Sendryx closes all three gaps.

01

Runaway spend

Agents loop. Agents retry. Without hard budget caps, a single bug becomes a five-figure bill. Cost Control gives every namespace a ceiling — requests return 402 the moment you hit it.

02

Leaked secrets

Agents pass API keys in request bodies. One misconfigured tool and your keys are in a log file. The Gateway scrubs secrets from every outbound request before they leave your stack.

03

No memory

Agents restart cold every session. They repeat questions, forget preferences, re-fetch data they already have. Memory API gives agents persistent, searchable context — across sessions, across agents.

Cost Control

Set budgets. Get alerts. Agents stop when the money runs out — not after.

⚙️

Budget Configuration

Set daily, hourly, and total spend caps per namespace. One API call to lock in your limits across all agents and models.

Automatic Circuit Breakers

When a namespace hits its limit, requests return 402 instantly. No runaway spend, no surprise invoices at end of month.

🔔

Webhook Alerts

Get notified at 80% and 100% of budget via webhook. Slack, PagerDuty, or your own endpoint — configure once and forget.

Live spend dashboard

Cost by namespace, by model, by hour. Compare daily vs hourly burn. Spot cost drivers before they become a problem.

Open Dashboard →

One proxy. Every agent call.

Rate limiting, secret detection, and full observability built in. Every AI call through a single auditable chokepoint.

🔀

Transparent Proxy

Forward any HTTP request through the gateway. Your agent's API calls land here first — route to OpenAI, Anthropic, or any target you specify.

🚦

Rate Limiting

Hard caps per API key: 100 req/min and 1,000 req/hr. Runaway agents hit 429s — not your credit card limit.

🔒

Secret Scrubbing

Outgoing request bodies are scanned for API keys, tokens, and passwords before forwarding. Secrets get redacted. Nothing leaks.

// Route every agent call through the gateway — any framework, any LLM const response = await fetch('/v1/gateway/proxy', { method: 'POST', headers: { 'X-API-Key': 'sk-gw-...' }, body: JSON.stringify({ target_url: 'https://api.openai.com/v1/chat/completions', method: 'POST', headers: { 'Authorization': 'Bearer [REDACTED]' }, // auto-scrubbed body: { model: 'gpt-4o', messages: [...] } }) }); // { success: true, status_code: 200, latency_ms: 312, secrets_detected: false, response: {...} }

Full observability dashboard

Every proxied request by date, target URL, or status code. Latency trends, error rates, top API targets. Understand exactly where your agents are spending time.

Open Gateway →

Memory API

Persistent memory for AI agents — store, search, forget. One API. Agents that actually remember.

💾

Store

Write any text to a namespace-scoped memory store. Embeddings are generated automatically — no vectors to manage. Attach metadata. Set a TTL for short-term memory that auto-expires.

🔍

Semantic Search

Retrieve by meaning, not keywords. Cosine similarity search across your namespace with a configurable relevance threshold. Returns ranked results with similarity scores.

🗑️

Forget

Delete a single memory by ID or wipe an entire namespace. Explicit deletion only — nothing disappears unless you ask. Useful for user data erasure and compliance workflows.

// Store a memory — embedding generated automatically await fetch('/v1/memory/store', { method: 'POST', headers: { 'X-API-Key': 'sk-gw-...', 'Content-Type': 'application/json' }, body: JSON.stringify({ content: 'User prefers dark mode and weekly summary emails', metadata: { user_id: 'u_123', type: 'preference' }, expires_in_seconds: 86400 // optional TTL — omit for permanent memory }) }); // → { success: true, memory: { id: 42, namespace: "prod", created_at: "...", expires_at: "..." } } // Semantic search — finds related memories by meaning const results = await fetch('/v1/memory/search?query=user+display+settings&threshold=0.4', { headers: { 'X-API-Key': 'sk-gw-...' } }).then(r => r.json()); // → { results: [{ id: 42, content: "...", similarity: 0.87, metadata: {...} }] }
⏱️

TTL Support

Set expires_in_seconds on any memory. Expired memories auto-purge hourly. Short-term vs long-term context, built in.

📊

Usage Tracking

Track stores, searches, deletes, and storage bytes per namespace via GET /v1/memory/usage. Know your memory footprint.

🔐

Namespace Isolation

Every API key scopes to its own namespace. Cross-namespace reads are impossible by design — your agents' memories stay separate.

🚦

Rate Limited

60 requests/min, 600/hr per key on memory endpoints. Prevents runaway embedding generation costs from misbehaving agents.

Full API documentation

5 endpoints: store, search, forget, list, stats, usage. Curl examples, response schemas, error codes.

View Docs →

Per-agent pricing. No surprises.

Pay for what you run. Gateway, Cost Control, and Memory API included on every plan.

Starter
$99/mo
per agent
For teams shipping their first production agents.
  • Up to 5 agents
  • Full Gateway access
  • Cost Control + budget caps
  • Memory API (when available)
  • 30-day request history
  • Webhook alerts
Get Started Free

Frequently asked questions

Yes. Sendryx is framework agnostic — it works as an HTTP proxy, so it's compatible with LangChain, CrewAI, the Anthropic Claude Agent SDK, OpenAI Agents, AutoGen, and any stack that makes HTTP requests. No SDK installation required to use the Gateway or Cost Control.

Provider-side limits apply to the whole account, not per agent or namespace. Sendryx lets you set granular caps per team, campaign, or workflow. When a namespace hits its limit, we return 402 immediately — before the request goes upstream — so you're never charged.

Memory API is live. Store any text using POST /v1/memory/store — we auto-generate embeddings (OpenAI text-embedding-3-small, 1536 dims). Search semantically with GET /v1/memory/search. Delete individual memories or wipe a namespace. TTL support lets memories auto-expire. All memories are namespace-isolated by API key.

Every outbound request body is scanned with pattern matching for common secret formats — API keys, bearer tokens, connection strings, private keys. Detected secrets are redacted before the request leaves Sendryx, and you're flagged via webhook. Nothing reaches the target URL.

Yes. All plans are month-to-month. No contracts, no cancellation fees.

Live Demo

See semantic memory in action.

No account needed. Store a sentence, search with different words — watch the match happen.

1 Store

Store a memory

Any text. Embeddings generated automatically.

curl -X POST /v1/demo/store \ -d '{"content": "The dog ran fast", "session_id": ""}'

Search with different words

Semantic, not keyword. "quick canine" finds "dog ran fast".

curl /v1/demo/search\ ?query=quick+canine&session_id=
✓ Semantic match! "quick canine" matched "The dog ran fast" — no shared words, just meaning.
Get your API key instantly
Free to start · no credit card · key in your inbox too

Stay ahead of the curve.

Get Sendryx release notes, agent infrastructure insights, and updates on new features. No spam — one email per week max.

Agents fail in predictable ways. Fix all three.

Memory. Cost control. Gateway. Every production agent stack needs all three — and right now, no one else ships them together. Sendryx does.

Get Your API Key Free