No pre-built connectors to wait for. If a system has a REST or HTTP API, Samba can integrate it.

• Any REST / HTTP endpoint — API keys, bearer tokens, basic auth, and two- and three-legged OAuth, configured once and reused across flows
• JSON and XML sources — XML feeds are ingested natively and converted for mapping alongside JSON
• Synchronize (upsert) — compares source and target, inserts what is new, updates what changed
• Push (insert-only) — sends every source record without reading the target
• Webhooks — processes inbound event payloads in real time and routes them to insert or update on a key match
• Multi-step flows — chains lookups and writes across several systems in one run; each step can use the source data and every prior step's response
• Enrichment — fetches related data from a second system before writing to the target
• Write-back — stamps a returned value, such as the target's new ID, back onto the source record
• Composite matching — matches records on several key fields at once
• Incremental sync — fetches only changed records via a timestamp parameter in the format you choose; the cursor advances only after a successful run
• Pagination — follows next-link and offset/page cursors until the full record set is retrieved
• Python transforms — field mappings are plain Python expressions — no proprietary mapping language

AI drafts; a deterministic grammar decides whether the draft is admissible. The platform also runs with AI switched off.

• Natural-language builder — describe the goal in plain text and the AI drafts the full integration plan
• System and entity discovery — identifies source, target, and enrichment systems from your description
• Spec and payload analysis — reads a pasted OpenAPI spec or sample payload and extracts its fields
• Endpoint and key-field suggestions — proposes HTTP method, URL pattern, list path, and the likely matching field, for you to confirm
• Python transform suggestions — drafts field-level mapping expressions you can read and edit
• Conversational refinement — follow-up instructions revise roles, operations, and lookup chains
• Grammar-validated output — every plan is checked against a deterministic grammar before it can be saved or run
• Configurable model endpoint — model, endpoint, and request shape are set per deployment and change without a redeploy; Azure OpenAI today

The same discipline pointed at failures: the AI selects from a closed catalog and labels its own certainty.

• Execution failure analysis — pinpoints the failed stage, whether it is transient or permanent, and the responsible mapping or request
• Deployment diagnostics — analyzes pods, database, Vault, scheduling, run history, and a live Kubernetes/Flux snapshot
• Closed finding catalog — findings carry stable, versioned codes; the AI cannot invent a new one
• Certainty labels — every finding is marked confirmed, likely, possible, or ruled out
• Evidence anchors — each finding cites the evidence behind it and expands to the raw data
• Guided remediation — signed, scoped recovery actions with a rollback path, admin confirmation, and an audit entry — never free-form commands
• Secret-scrubbed bundles — metadata, counts, and status only — no payloads, secrets, or customer data — safe to attach to a ticket

Compliance capability ships as the platform. It is not an add-on and not priced separately.

• Single-tenant, in your cluster — deploys into your cloud account and network; no shared runtime with any other customer
• Isolated executors — production runs separately from dev/test, with exclusive access to production credentials
• Bring your own IdP — OpenID Connect with Entra ID, Okta, Google, or any compliant provider
• Secrets management — credentials held in HashiCorp Vault or Azure Key Vault, never in config or images
• Tamper-resistant audit log — every change records who, when, and old/new values; immutability is enforced by the database
• Live control evaluation — 54 security controls evaluated against the running cluster, each mapped to SOC 2 criteria and NIST 800-53
• Restore-verified backups — daily backups, an actual restore every week, and a control that fails if the last good restore is over seven days old
• Promotion with approval gates — Development → Testing → Production, with integrations exportable as JSON templates for review and reuse
• Monitoring and alerting — scheduled or on-demand runs, step-by-step logs, structured errors, configurable alerts