Connectors
Connect your HTTP systems.
Keep every call governed.
Turn an existing endpoint into an agent tool, a pre/post-turn state bridge, or a signed outbound event webhook. Test it live, attach it to bots, and trace what happens in production.
3
runtime connector types
7
authentication modes
5
supported HTTP methods
4
lifecycle states
Connectors work with customer-provided HTTP endpoints. They are not a catalog of prebuilt vendor integrations or direct database drivers.
From endpoint to production
A connector is more than a URL.
Define how an agent may use your HTTP service, verify the real exchange, scope it to the right bots, and keep its production behavior visible.
One connector registry, three runtime paths
loaded fresh for every turn
Conversation turn
A user message enters the governed agent runtime.
Your HTTP service
You own the API or webhook consumer; Rylvo governs the exchange.
Define
Choose Tool, State Sync, or Event; then describe the endpoint, method, timeout, and type-specific behavior.
Authenticate
Select one of seven auth modes. Sensitive values are write-only in the UI and encrypted before storage.
Save & test
Create the connector, then run a live endpoint test with the saved auth, transforms, timeout, and retry policy.
Attach
Scope the connector to the bots and workflow stages that need it. Tool approval requirements travel with it.
Activate
Active connectors are loaded from the registry on every turn; State Sync and Event connectors run automatically.
Observe
Inspect tests, call outcomes, delivery logs, health signals, dead letters, and connector context inside traces.
Three runtime roles
Call, synchronize, or notify.
Each connector type has a specific place in the turn lifecycle. Choose one job per connector so its behavior stays explicit.
Let the model call your API
Tool connector
Turn an existing HTTP endpoint into a named function the model can choose during a conversation.
GET, POST, PUT, PATCH, or DELETE
Model-facing input schema and simple request/response field maps
Workflow-stage scope, approval gates, idempotency, retries, and circuit breaking
Bring remote context into each turn
State Sync connector
Read state from your HTTP service before a turn and write collected business facts back after it.
Read, write, or bidirectional modes with separate endpoints
Field mapping plus patch or replace writes
Fail-open reads and If-Match optimistic concurrency support
Notify the rest of your stack
Event connector
Send signed outbound webhooks after live workflow decisions and tool executions.
Outbound delivery only—not an inbound webhook receiver
Signed payloads, event IDs, retry policy, and delivery history
Dead-letter inspection, replay, and signing-secret rotation
Authentication & boundaries
Use the auth your endpoint expects.
Seven authentication modes cover public endpoints, internal services, signed requests, and machine-to-machine OAuth—without putting secrets in a prompt.
None
Public or network-protected endpoint
API key
Custom key header
Bearer token
Authorization bearer header
Basic auth
Username and encrypted password
HMAC-SHA256
Timestamped request signature
OAuth2
Client credentials with token caching
Custom headers
Your required header set
Encrypted, write-only secrets
API keys, bearer tokens, passwords, HMAC secrets, and OAuth client secrets are encrypted at rest and masked on read.
Safer outbound networking
Endpoint validation blocks private-address SSRF paths, redirects stay disabled, and responses are capped at 1 MB.
Managed machine tokens
OAuth2 client-credentials tokens are cached; a 401 invalidates the cached token so a retry can obtain a fresh one.
Signed event delivery
Outbound events carry a signing secret independent of endpoint auth, with controlled secret rotation and overlap.
Production operability
See the failure. Limit the blast radius.
A working demo is not enough. Connector operations expose what was sent, what came back, what failed, and what an operator can safely do next.
Test the real request
The saved connector can make a live HTTP test so you can inspect response status, body, headers, and latency before activation.
Bound failure
Timeouts, per-connector rate limits, exponential backoff with jitter, response-size limits, and circuit breakers constrain bad dependencies.
Watch health
Periodic probes and consecutive-failure counts update connector health. Operational status remains an explicit operator decision.
Investigate outcomes
Use call and delivery records, last-test results, success and failure totals, average latency, and canonical traces to find problems.
Replay deliveries
Failed outbound events land in a dead-letter view and can be replayed after the receiving endpoint is repaired.
Roll back configuration
Restore an earlier endpoint, retry, mapping, scope, or type configuration without erasing history. Secrets and health are intentionally separate.
Runtime wiring
The registry remains the source of truth.
Active connectors are loaded fresh on every engine turn. Tool schemas, stage scope, and approval rules arrive together. State Sync and Event connectors run as automatic hooks, never as tools the model can improvise with.
Plan capacity
Every plan can connect.
Connector count is enforced server-side when a connector is created. Upgrade for a larger registry; the three connector types share the same pool.
Free
1
connector
Lite
15
connectors
Pro
50
connectors
Team
50
connectors
Enterprise
Unlimited
connectors
Need to compare connector capacity with bots, skills, knowledge sources, channels, and other platform limits?
View full pricingClear expectations
What Connectors do—and do not do.
The short answers teams need before they design an integration.
Are these native integrations with specific vendors?
No. Rylvo Connectors are governed custom HTTP connectors. Your API, system of record, or webhook consumer must expose a reachable HTTP endpoint. Rylvo does not provision that API or connect directly to a database socket.
Do Event connectors receive inbound webhooks?
No. Event connectors deliver outbound, signed webhooks from Rylvo to your endpoint. The current live engine hooks cover completed workflow decisions and completed tool executions.
How are tool schemas used?
A Tool connector’s input schema describes the function arguments presented to the model. Request and response field maps can reshape payloads. State Sync reads can separately validate returned state against a configured schema. Rylvo does not claim universal JSON Schema validation of every tool request and response.
What happens when I attach a connector to a bot?
Active linked connectors are loaded fresh for each turn. Tool connectors become callable functions with their scope and approval policy. State Sync and Event connectors remain automatic runtime hooks. When an eligible response-composer prompt exists, Rylvo also maintains a connector usage block as a versioned prompt edit.
Can I test before activating?
Yes. Save or create the connector first, then run its live test from the dashboard before changing it to Active. Testing does not activate a connector automatically.
What does connector rollback include?
Version snapshots cover connector configuration such as endpoints, retry policy, type-specific settings, bot and workflow scope, and tags. Credentials, health, metrics, test results, and lifecycle status are deliberately not restored from a configuration version.
Your endpoint. Governed by Rylvo.
Give your agents access without giving up control.
Define one HTTP connector, test the real exchange, attach it to the right bots, and keep its production behavior visible.