The unified AI abstraction and aggregation layer — now with self-evolving agent skills (SkillOpt), in-process ML inference (ML.NET), DID/Verifiable Credential identity, karma-gated AI and full ACP/ANP/gRPC/GraphQL protocol support. Built on WEB4 and WEB5, routing intelligently across 20+ providers.
A complete 25-point architecture upgrade — from self-evolving AI skills and in-process ML to DID identity, multi-protocol orchestration and karma-gated access. This is the platform investors and enterprise teams have been waiting for.
SkillOpt (Microsoft Research, arXiv:2605.23904) — each FAHRN agent evolves its own best_skill.md procedure through a rollout → reflect → edit → gate loop. +23.5% avg gain demonstrated. No weight updates — skills transfer across models and harnesses. Paired with ML.NET in-process AutoML at microsecond latency.
ACP (BeeAI/IBM), ANP (DID-based discovery), gRPC bidirectional streaming, GraphQL subscriptions, and AsyncAPI/Kafka event mesh — all behind a single unified endpoint. FAHRN agents can now be discovered, invoked and coordinated across any agent runtime on the internet.
W3C DID + Verifiable Credential support (did:key, did:web, did:ethr, did:ion). AI agents act on behalf of avatars via cryptographically-signed VC grants. Every FAHRN dispatch is attributable to a provable, consent-governed identity — critical for enterprise compliance.
Avatar karma tiers gate model access and monthly token budgets — Bronze through Diamond. Karma earns through contributions, OAPPs, community and verified credentials. Creates a self-sustaining incentive loop where high-quality participation unlocks more powerful AI capabilities.
External memory provider integration (Mem0, Zep, Letta, LangMem, Graphiti, Redis Vector), upward memory propagation API, and WebSocket bidirectional sessions with persistent server-side state in Holonic BRAID — turning every conversation into a living memory artifact.
OpenTelemetry traces and Prometheus metrics on every FAHRN dispatch, memory operation and provider call. Grafana dashboards, structured JSON logging, cost-per-dispatch analytics and real-time alerting — production-grade telemetry from day one.
The AI landscape is fragmented — dozens of providers, incompatible APIs, competing agent protocols and orchestration frameworks. WEB6 is the abstraction layer that unifies them all, so you build once and deploy everywhere.
WEB6 normalises the request and response interface for every major AI provider and open-source model runtime. Switch providers with a single config change — no refactoring required.
More than a proxy — WEB6 is an intelligent orchestration and normalisation layer sitting on top of both WEB4 and WEB5. Every AI request is grounded with real-world identity, cross-reality app data and a living holon graph — making it the most contextually rich AI API ever built.
Authenticate with your OASIS avatar key and start making requests immediately — no per-provider SDK setup, no credential juggling.
// Route a completion to whichever model fits best POST https://api.web6.oasisomniverse.one/v1/complete Authorization: Bearer <your-oasis-avatar-key> { "provider": "auto", // "auto" | "openai" | "anthropic" | "gemini" | "groq" | "mistral" | ... "model": "auto", // or a specific model ID "messages": [ { "role": "user", "content": "Hello from the OASIS" } ], "routing": { "priority": "cost", // "cost" | "latency" | "quality" "fallback": true }, "avatarId": "your-avatar-id" // optional — injects WEB4 identity/karma context }
// JavaScript / Node.js — install the npm package npm install @oasisomniverse/web6-api import { Web6Client } from '@oasisomniverse/web6-api'; const web6 = new Web6Client({ baseUrl: 'https://api.web6.oasisomniverse.one' }); web6.setToken('<your-oasis-jwt>'); const { result } = await web6.completion.complete({ Provider: 'auto', Model: 'auto', Messages: [{ Role: 'user', Content: 'Hello from the OASIS' }], UseFAHRN: true, // optional: FAHRN reasoning pipeline UseHolonicBraid: true, // optional: shared graph injection Routing: { Fallback: true }, }); console.log(result.Content);
// Dispatch a problem straight to FAHRN - the controller agent scores agents, picks a // dispatch mode, runs loop detection and returns the assembled Mermaid execution plan POST https://api.web6.oasisomniverse.one/v1/reasoning-network/dispatch Authorization: Bearer <your-oasis-avatar-key> { "problem": "Plan a zero-downtime migration of a 50M row table", "taskType": "architecture", "mode": "Parallel" // "Serial" | "Parallel" | "Decomposed" }
WEB6 doesn't just abstract AI providers — it sits on top of the entire OASIS data stack. WEB4 and WEB5 feed their world data directly into the AI layer, making every request aware of identity, reputation, cross-world state and a living ecosystem of apps and holons.
The universal identity, reputation and storage unification layer. Every AI request in WEB6 can be grounded with your avatar's full history, karma score and COSMIC ORM data across 40+ storage providers — giving AI models context that no other platform can offer.
STAR ODK and STARNET publish every app, asset and interaction as a versioned holon in a living graph. WEB6 pulls this entire cross-world application data stream — all OAPPs, user interactions and holon state — directly into the AI layer, making it extraordinarily context-aware.
WEB6 is the only AI abstraction layer that feeds real-world identity and cross-chain data (WEB4) and a live ecosystem of cross-world apps and holons (WEB5) directly into every AI model call — making it the most powerful and contextually rich AI API ever built.
OASIS extends the OpenSERV BRAID framework (Amçalar & Cinar, arXiv:2512.15959) with a fractal holonic shared graph library — delivering 74× PPD gains on GSM-Hard that scale indefinitely across agents and sessions. One shared library. Every reasoning graph ever generated — available to everyone, forever.
FAHRN is the self-optimising reasoning mesh built on top of Holonic BRAID. A controller agent dispatches work to scored agents in serial, parallel or decomposed mode — using a composite score of category performance, speed, cost and loop-detection stability — and learns from every outcome via EMA score updates stored in the holonic memory hierarchy.
The controller dispatches the problem to the highest-scoring agent for that category. If it stalls, the next best agent takes over. One agent works at a time — lowest token cost, still intelligent routing.
All agents receive the same problem simultaneously. Each produces a Mermaid execution diagram. The controller compares plans — selecting the best, or merging complementary diagrams into a superior combined plan.
Large, complex problems are broken into sub-problems. Each is routed to the agent best matched to its category and speed profile. The resulting sub-diagrams are then combined by the controller into one unified execution plan.
best_skill.md artifact — a natural-language procedure document that evolves through a rollout → reflect → edit → gate loop, without touching model weights. Demonstrated +23.5% average gain across benchmarks. Skills transfer across models, harnesses and task categories — stored as holons in Holonic BRAID.Connect any MCP-compatible AI agent or tool directly to the full OASIS platform — avatar unification layer, karma, COSMIC ORM data and the entire WEB6 AI layer — through a single standardised Model Context Protocol interface.
The MCP server also runs as an HTTP endpoint built into the WEB6 REST API — for cloud-hosted agents, claude.ai connectors, and any client that cannot use stdio:
GET /mcp — SSE stream (HTTP MCP transport) POST /mcp — MCP JSON-RPC tool calls GET /.well-known/mcp.json — auto-discovery document GET /.well-known/agent.json — A2A agent card
These endpoints are not shown in the Swagger UI — they use SSE/streaming transports the interactive explorer cannot represent. Point your MCP client at the discovery URL and it configures itself automatically. Full details in the MCP Server README.
A desktop coding environment built around the same WEB6 AI layer — agentic coding assistants (Claude, OpenServ and every model in the SERV catalog), MCP tool access and full OASIS/STAR project awareness, in one editor.
The AI infrastructure market is fragmenting at exactly the moment enterprises need consolidation. WEB6 is the only platform that unifies the entire stack — from models to memory to identity — and already has 8 years of production infrastructure underneath it.
WEB6 is not a wrapper around OpenAI. It is an intelligence operating system — with its own reasoning network (FAHRN), its own shared memory layer (Holonic BRAID), its own identity system (WEB4 karma + DID/VC), and a data moat built over 8 years that no new entrant can replicate. The protocol-neutral architecture means it gets stronger with every new AI model that launches — each new provider is another node in the network.
WEB6 v2.0 is fully operational — all 25 improvements across Phases 1–3 are shipped as of July 2026. Phase 4 is the long-horizon AGI pathway.
POST /v1/fahrn/solve — single hero endpoint: problem in, reasoned answer + Mermaid plan + telemetry out@oasisomniverse/web6-api npm + oasis-web6 Python packageGET /v1/context/avatar/{id} rich context block for any external systembest_skill.md holons; demonstrated +23.5% avg gain (Microsoft Research, arXiv:2605.23904)