Beta · Early Access

For CrewAI, AutoGen, LangGraph, and A2A protocol implementers

When AI agents from different vendors collaborate, who do you trust?

Memory poisoning, identity spoofing, and consensus collapse all cross vendor boundaries. CrewAI agent calls an AutoGen agent calls an in-house agent — and now the trust assumption is whichever one is weakest. There is no shared identity layer for AI agents today.

Sgraal's cross-agent provenance tracking, a Python SDK + MCP server, and 6 published framework adapters are production (3 more in beta). The A2A Treaty Engine — a trust handshake mediator for cross-vendor agent collaboration — is in Beta. Multi-agent platform teams shape the protocol with us.

Important — compliance evidence: /v1/check verdicts are not included in the audit trail and do not generate W3C Verifiable Credentials. For compliance-grade evidence (HIPAA, GDPR, EU AI Act, FDA 510(k), NIST AI RMF), use /v1/preflight, which produces full audit-log entries and signed W3C VCs per verdict. The /v1/check endpoint is suitable for experimentation and high-frequency agent gating only.

What works today

Four primitives every multi-agent system needs.

Production: the Python SDK, MCP server, and 6 published framework adapters (3 more in beta).

widgets

Python SDK + MCP server + 6 published adapters (3 beta)

Live

Native integration for LangChain, CrewAI, AutoGen, LlamaIndex, Haystack, Semantic Kernel, OpenAI Agents, Mem0, and 15 more. Each SDK is a thin shim over the canonical Python client — same scoring engine, framework-idiomatic API.

hub

MCP server for Claude Desktop

Live

@sgraal/mcp drops Sgraal into Claude Desktop. Any conversation can call sgraal.preflight on memory state directly. The reference MCP integration for AI safety.

account_tree

Cross-agent provenance tracking

Live

Every memory entry carries a signed provenance chain. When agent A passes memory to agent B, the chain extends. Loops, drift, and identity-laundering across vendor boundaries become visible.

stream

SSE streaming for multi-agent debugging

Live

/v1/preflight/stream emits 23 events across 4 phases per decision. For multi-agent systems, this gives per-agent and cross-agent visibility into how a fleet decision was formed.

What is in active development

Two Beta features for multi-agent platform design partners.

In active development with design partner input. Not yet GA — but real, not vaporware.

handshake

A2A Treaty Engine

Beta

Google's A2A (Agent-to-Agent) protocol does not specify trust mediation between agents from different vendors. The Treaty Engine adds Subjective Logic + Provenance + Trust enum translation as an explicit handshake step. Agent A claim is rewritten into agent B's trust framework before consumption.

Status: core Subjective Logic and Provenance pieces are GA; the A2A-protocol-compatible handshake adapter is in test. Design partners get the first integration and influence the protocol-extension shape.

vaccines

Cross-agent vaccine propagation

Beta

Sgraal already does fleet-wide vaccine propagation within one tenant. The cross-agent extension propagates attack signatures across multi-agent meshes — one agent flagged for poisoning blocks the same signature on sibling agents within seconds. Beta Federation / cross-agent propagation is in active development — not yet production-ready. Early access: hello@sgraal.com.

Status: in-tenant vaccine propagation is GA. The cross-mesh extension is in design with partners — the privacy and trust boundary between meshes is the open question.

Design Partner Program

First 5–10 multi-agent platform architects shape the roadmap.

Early Access is not a feature flag — it is a structured program. Design partners get founder-direct access, priority on roadmap input, and locked-in pricing through the Beta period.

connect_without_contact Founder-direct access

Slack channel with the founder. No ticket queue, no AE filter. You hit a wall, you hear back same day.

route Roadmap input

Beta features ship based on what design partners actually use. The next Beta milestone is decided by the first 5–10 customers, not a marketing committee.

lock Locked-in pricing

Your Beta-period pricing is locked through general availability. If the GA tier list moves up, you stay on the partner rate for 24 months.

handshake Honest reciprocity

In return: candid feedback when something does not work, a written case study at GA (with your approval and disclosure terms), and a willingness to live with Beta-grade rough edges.

Apply to the program →

No NDA required to apply. We respond within 48 hours.

Beyond Beta

Where this is heading.

Per-handshake pricing tier

Concept

Long-horizon commercial model: per-A2A-handshake pricing as a separate tier from per-decision. Floated as a direction; structural pricing work has not started.

More on the long-term direction →
info

Honest disclosure: what works today, what is beta.

  • Today (Live): Python SDK + MCP server + 6 published framework adapters (3 in beta), cross-agent provenance tracking, SSE streaming for multi-agent debugging.
  • Beta: A2A Treaty Engine (Subjective Logic + Provenance enum handshake adapter); cross-agent vaccine propagation. Core pieces GA; the cross-vendor protocol layer is what's in test.
  • Concept: Per-handshake pricing tier. Long-horizon commercial design, not built.

Tier badges across the page surface this distinction. No category was glossed over.

Five to ten design partners. That is the cohort.

If you are a multi-agent platform architects and want to shape what comes next, write us. If you would rather wait for GA, the page is here when you are ready.