Route Claude Opus traffic to Sonnet
Claude Opus handled 96 calls in this range. Claude Sonnet reaches similar quality on most of this traffic.
Workflows · Agents · AI Router
Build workflows on a visual canvas, run agents with memory and retrieval, keep a human in the loop — and route every model call to the cheapest model that does the job.
Self-hosted · Your keys · Your data
The problem
Cost
“Why did costs increase?”
Reliability
“What happens when a provider goes down?”
Control
“Who approved this action?”
Context
“Where does the agent’s knowledge come from?”
DirectOS helps answer every one of them.
Execution lifecycle
One loop from the first node to the last optimization — the same lifecycle that structures the product itself.
01
Compose workflows on a visual canvas — triggers, logic, code and AI in one graph.
02
Run them on a schedule, a webhook or a click — with retries, approvals and every node streaming live.
03
See what ran, what it cost and what to change. Then change it.
Platform
Agents, memory, retrieval and human approval are part of the engine — not plugins you wire together.
DirectAgents
Give an agent a goal and a set of nodes. It plans, calls them as tools, reflects — and delegates to other agents.
delegate_to_agentmessage_agentDirectMemory
Conversation and vector memory on pgvector, summarized automatically and expired when it goes stale.
memory.storememory.recallRAG
Ingest PDF, DOCX, HTML and CSV. Hybrid search fuses vector similarity with full-text ranking.
rag.ingestrag.queryHuman in the loop
An approval node pauses the run until someone approves or rejects it — with a Slack or Discord ping.
logic.approvalIntegrations
Postgres, Slack, Telegram, GitHub, email, Google Sheets and S3 — credentials sealed with AES-256-GCM.
postgres.queryslack.sends3.objectLLM Gateway
An Anthropic-compatible API with per-project keys, budgets and fallback chains — every call logged and priced.
POST /v1/messagesAI Operations Center
The Operations Center ships with the platform. These are its instruments — real UI, real signals.
01Money Saved
02Provider Health
03AI Recommendations
04Router Insights
05Execution Timeline
06Cost Analytics
AI Router
Every task is scored for complexity, then routed to the cheapest model that clears the quality bar — with ordered fallbacks when a provider degrades.
The right model for every task.
Features
From the first workflow to production operations — one surface, no glue code.
Workflow Builder
A visual canvas for triggers, logic, code and AI — with undo, copy and paste.
Versioning
Every save is a snapshot. Restore any version, import and export as JSON.
Sub-workflows & Loops
Call a saved workflow from another, or run it once per item in a list.
Scheduler
Cron triggers with time zones, alongside webhooks and manual runs.
Distributed Workers
Run in-process by default, or fan executions out over Redis-backed queues.
AI Router
Complexity-aware model selection with ordered fallbacks and cost accounting.
Ensembles
Send one prompt to several models — first success, all, or majority vote.
Benchmarks
Measure latency, cost and success rate of models on your own prompts.
AI Operations
Spend, savings and provider health across workflows and the gateway.
Accounts & Roles
Viewer, editor and admin roles, API keys and an audit log of sensitive actions.
Providers
Ten providers — hosted frontier models, your own cloud and local weights — behind one routing surface.
Roadmap
No fake doors — a feature appears here before it appears in the product.
Shipped
Workflow engine & Studio
Visual editor, versioning, sub-workflows, loops, scheduler, approvals, 25 built-in nodes.
AI Router v1
Ten providers, complexity scoring, fallbacks, ensembles, benchmarks and an LLM gateway.
Agents, memory & RAG
Tool-calling agents, pgvector memory and hybrid document search.
Teams & security
Accounts, roles, API keys, audit log, encrypted credentials, distributed workers.
Next
AI copilots in Studio
Generate workflows from a sentence, explain them, debug failed runs.
Marketplace & SDK CLI
Publish and install nodes, templates and agents.
Observability
OpenTelemetry traces, Prometheus metrics, cost forecasts.
Later
SSO & organizations
OAuth2/OIDC sign-in and multi-tenant workspaces.
DirectCloud
Managed hosting with billing, quotas and usage metering.