| The client | Requests, context windows, token optimization, cost checks, reasoning, prompt caching, streaming |
| The lifecycle | Authorization, a shared budget, hooks, audit, and metrics for every call of every family |
| Providers and routing | The twelve completion providers, routing and failover, the model registry |
| Guardrails | Injection, PII, secrets, schema validation, output redaction |
| Graphs | Parallel branches, interrupts, checkpoints, time travel, subgraphs, caching, diagrams |
| Agents and tools | Graph-based agents with approvals, capabilities and permission policies, a sandbox contract, the middleware catalog, and the simple tool loop |
| The deep-agent preset | Long, multi-step agents: a plan, a sandboxed workspace, helpers, skills, and a context that stays small |
| Long-term memory | Namespaced memory with semantic search, in memory or Redis |
| MCP | Borrowing tools from MCP servers, a registry of many, and serving your own |
| Agent protocols | Serving an agent to user interfaces over AG-UI, to other agents over A2A, and to code editors over ACP, and calling remote agents as tools |
| Traces | Run trees, queries, feedback, comparison, and alerts |
| Evaluation | Datasets, evaluators, experiments, comparisons with a verdict, review queues, LLM judges, and tools emulated by a model so an evaluation sends nothing |
| Context hub | Prompts, instructions, tools, and skills versioned together, promoted through gates, and moved between projects |
| Insights | Failing and slow runs clustered, regressions between time windows, and evaluated fix proposals |
| Prompts | Typed templates, content versions, gated promotion, rollback, A/B splits, serving through outages |
| Postgres | One adapter family for operations, memory, traces, evaluation, circuits, and prompts |
| SQLite | Durable operations, checkpoints, and memory on one machine, over any SQLite driver |
| Durable operations and jobs | Operations that survive a restart, webhooks, and in-process job helpers |
| Resilience and observability | Circuit breaking, distributed rate limits, metrics, logs, and audit |
| Batch tiers | The providers' discounted batch APIs, resumable from any process |
| Caching | Exact and semantic response caches, with memory, Redis, and SQLite adapters |
| Grounding | Retrieval, citations, verification, self-consistency, and knowledge graphs |
| Loaders | Files, Markdown, HTML, CSV, JSON, PDF, web pages, sitemaps, and Git repositories into retrieval |
| Retrieval | Hybrid search, reranking, MMR, parent documents, and Redis, Pinecone, Weaviate, and Chroma stores |
| Embeddings and retrieval | Embeddings as a routed operation family, and RAG helpers |
| Media understanding | Reading documents and images page by page, describing images, and answering questions about them, through any vision model or OCR |
| Typed decisions | A yes or no, a choice, or a score, answered with probabilities and a calibrated confidence, from TypeSafe AI's Jev or any model, so a graph, an approval, or a guardrail acts when it is sure and asks a person when not |
| Video generation | Videos from a prompt or a first frame, on Google Veo and Runway, as durable jobs streamed into an asset store and priced per second |
| Images (experimental) | Generation and masked edits over three backends, input safety, moderation, asset stores |
| Voice | Transcription, speech, voice turns, and voice sessions |
| Realtime voice | Browser and server realtime sessions with barge-in, tools, and exports |
| Telephony | Calls, webhooks, phone numbers, and phone agents on realtime sessions |
| Testing | Recording and replaying provider traffic, and conformance suites |
| Adapter kit | Building a provider, store, or retriever outside this repository, verifying it against the same contracts, and its compatibility record on the compatibility page |
| Studio | A shared UI for traces, threads, approvals, experiments, prompts, deployments, costs, and health, as a separate package |
| Agent server | Self-hosted HTTP server for assistants: threads, durable runs, resumable streams, cron, a worker queue, and metrics |
| Deployments | Revisions and canaries with automatic rollback, autoscaling on Kubernetes and Helm, and per-tenant limits |
| Tenancy | One tenant model: a tenant's view of every store, tenant graph runs, and the isolation suite |
| Policy | Guardrails, routing, budgets, residency, and credentials as one versioned file, validated as the client starts, checked by nexus policy check in CI, and recorded on every operation |
| Regions and credentials | Agent servers across regions behind a gateway, residency rules that keep each tenant's requests, traces, and assets in its regions, and each tenant's provider keys from a secret manager |
| Workflows | Ready-made chains and domain workflows |
| Command line | The nexus command |
| Packaging and install weight | Every entry point and what it costs |
| Runtimes | The portable kernel, and running the client, providers, graphs, and agents on Deno, Bun, edge runtimes, and browsers |
