Features

What ships when you embed Syncanix

Everything your agent can do — and the proof behind it. Every capability below is live and measurable in your own environment.

• Auto-discovered capabilities

  Point the CLI at your repo and it maps every endpoint into a typed action catalog — across ~29 frameworks in 12+ languages (Node, Python, PHP, Java, Go, Ruby, Rust, .NET and more). Extractor accuracy is F1 ≥ 0.85, then verified against live traffic by the runtime witness.

• Widget AND MCP from one catalog

  The same capability catalog drives an embeddable widget and a standards-compliant MCP server. Your customers chat from your app; their tools (Claude Desktop, Cursor, ChatGPT) connect over MCP. One source of truth, two distribution surfaces.

• Multi-provider from day one

  Anthropic, OpenAI, and AWS Bedrock-EU wired in on day one. Failover routing per request, cost-aware model selection, and no single-vendor lock-in.

• Hosted in the EU

  All customer data is stored in the EU (Frankfurt), and default AI inference runs there too — Claude on Amazon Bedrock in Frankfurt, never used to train a model. Retrieval stays in-region. Encryption at rest with per-tenant KMS keys, customer-held on Enterprise; EU-resident authentication; BYOK for your own LLM provider keys on every tier.

• Six languages, full RTL

  EN · ES · FR · DE · AR · HE at launch — and not just the UI: the agent’s answers and even its system prompts are localized. Arabic and Hebrew get true bidirectional layout with logical CSS, mirrored components, and native-speaker review every release.

• Audit trail by default

  Every capability invocation, every promote, every admin action writes to an append-only audit log. Mapped to AICPA Common Criteria CC6 (access) and CC8 (change) for SOC 2 evidence. Audit rows retained 13 months; security events copied to a 7-year write-once forensic archive.

• Eval-on-promote gate

  Every Dev → Prod promote runs your eval suite. Faithfulness, tool-call accuracy, and refusal accuracy regressions over 5 percentage points block the promote. Admin overrides leave a written justification in the audit trail.

• WCAG 2.1 AA, under 99 KB

  The embedded widget loads in under 99 KB gzipped, isolates inside a Shadow DOM, keyboard-reachable, screen-reader announced, AA contrast guaranteed. RTL and LTR render identically on mobile and desktop.

• Widget Design Studio

  Author your agent’s look in the Widget Design Studio — color, type, and shape mapped to your product. It auto-extracts a matching theme from shadcn, MUI, Chakra, and Mantine, so the widget is themed to your design system in minutes, not sprints.

• Acts as the signed-in user

  The agent acts as the signed-in user — never a shared bot with god-mode. The model never holds your users’ credentials, and every action is authorized and audited under that user’s own identity and permissions.

• Works with your IdP

  Plug into the identity you already run: Auth0, Clerk, Cognito, and WorkOS out of the box, plus generic OIDC, custom JWT, BFF session, and RFC 7662 introspection — 8 adapters in all.

• Guarded actions

  Side-effect confirmation gates and single-use HMAC action tokens stand between intent and impact, with RFC 9470 step-up authentication required before sensitive actions commit.

• Verify-intent SDKs

  Verify-intent SDKs in 8 languages — Node, Python, Go, Java/JVM, PHP, Ruby, Rust, and .NET — so your backend can verify every agent action natively, in its own stack.

• Drift detection

  The api-witness runtime middleware checks the discovered catalog against live traffic and flags drift — a moved route, a changed shape — before your users ever hit it.