Build memory-aware agents or plug your LLM directly into Virideon via MCP. Same brain that powers the Hub and the API.
API access is in private preview for design partners. The surface area below reflects what's in production today and stabilizing toward GA.
The Virideon MCP server is the fastest way to give your LLM real institutional memory. Configure it once; your model reads, searches, and persists memory through 90 production tools — with citations, provenance, and per-tenant isolation.
Request MCP access# Point any MCP-compatible client at your tenant AMBIENTOS_API_KEY=amb_xxx_yyy \ AMBIENTOS_API_URL=https://your-tenant.virideon.ai \ ambientos-mcp stdio
Every surface returns answers with citations to source, time, and person. Every call is scoped to your tenant. Every operation is logged.
A streaming agentic search with intent-aware profiles — executive briefing, fact checker, workflow automation, and more. Returns structured answers with citations to source, time, and person.
Four memory types under one surface — versioned semantic notes, immutable bi-temporal episodes, mutable state projections, and learned procedures. Each call traces back to the source it was learned from.
Documents, audio, email, and structured text. Virideon transcribes, diarizes, extracts entities, resolves identity, and places every event on a timeline — without you scripting any of it.
A live knowledge graph that knows who's who across documents, meetings, and emails — including via voice fingerprinting — and a relationship graph that surfaces dependencies and contradictions.
The actual request shapes you'd use as a design partner. API surface is stabilizing; design partners get the current reference and a heads-up on any breaking changes. Curl shown below — official Python and TypeScript clients coming with GA.
curl -X POST https://your-tenant.virideon.ai/api/v1/ai-search \ -H "Authorization: Bearer $AMBIENTOS_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "query": "What did the team commit to in the Q3 audit kickoff?", "profile": "executive_briefing", "stream": true }' # Returns an SSE stream of decomposition, tool calls, # synthesis, and a final cited answer.
import os, requests resp = requests.post( "https://your-tenant.virideon.ai/api/v1/ai-search", headers={"Authorization": f"Bearer {os.environ['AMBIENTOS_API_KEY']}"}, json={ "query": "What did the team commit to in the Q3 audit kickoff?", "profile": "executive_briefing", "stream": False, }, ) answer = resp.json()
const resp = await fetch( "https://your-tenant.virideon.ai/api/v1/ai-search", { method: "POST", headers: { "Authorization": `Bearer ${process.env.AMBIENTOS_API_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ query: "What did the team commit to in the Q3 audit kickoff?", profile: "executive_briefing", stream: false, }), } ); const answer = await resp.json();
curl -X POST https://your-tenant.virideon.ai/api/v1/agent-memory \ -H "Authorization: Bearer $AMBIENTOS_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "content": "Counsel firewall design pending confirmation as of May 12.", "source": { "type": "email", "id": "msg_..." }, "tags": ["q3-audit", "open-decision"] }'
curl -X POST https://your-tenant.virideon.ai/api/v1/ingest/document/text \ -H "Authorization: Bearer $AMBIENTOS_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "title": "Q3 Status Review", "content": "...", "source_type": "document" }' # Triggers async pipeline: chunk → embed → entity extraction # → identity resolution → graph insertion. Returns a job_id.
{
"thread_id": "thr_abc123",
"answer": "The team committed to launching the audit framework by Nov 15, with weekly checkpoints starting Oct 1.",
"citations": [
{
"source_id": "msg_xyz789",
"source_type": "email",
"timestamp": "2026-09-15T14:23:00Z",
"people": ["Sarah Chen", "Marcus Rivera"]
}
],
"intent": "decision_lookup",
"profile_used": "executive_briefing"
}
# Every citation traces back to source, time, and person.
# Streaming flows emit intermediate events before final synthesis.Every API key is bound to a tenant. Every request is logged. Sensitive operations require step-up auth — even for the holder of a write-scoped key.
Keys follow the pattern amb_{tenant}_{random} and are issued with explicit scopes. A key is bound to one tenant and cannot cross tenant boundaries.
Write paths and admin tools require a fresh re-authentication token via the x-reauth-token header. If the token is missing, the API returns a 403 with a reauth_required event for streaming flows.
API and MCP access are in private preview while the surface stabilizes. We work directly with each design partner — and pricing reflects the depth of that work.
If you're putting Virideon under your own agents, tools, or apps, request API access. We'll set up your tenant, issue scoped keys, and share the current reference.
Request API accessIf you want Claude, ChatGPT, or another MCP client to read and write Virideon directly, request MCP access. Quickest path to giving your existing AI tools real team memory.
Request MCP accessNot ready to apply? Join the GA waitlist →
30 minutes. We'll walk through your integration scenario and show you exactly what the API can and can't do today.
Usually within a few hours
We'll be in touch within a few hours with onboarding details.