The AI operating system for organisations
Your people and your AI, as one operational fabric
Your data, tools, and AI are scattered — and people do the integration in their heads. Oraclous forms a harness around all of it: one governed fabric where human members and AI Agents share task boards, hand work back and forth, and run under one model of control.
second mind — the unified operational fabric Oraclous forms around an organisation, where human members and AI Agents work side by side under the organisation's own access rules.
The fourth choice
Teams deploying agents today get three bad options.
Each trades away something you'll miss — speed, sovereignty, or sanity.
Bespoke code
Fast to start — brittle and one-off. Every change is an engineering ticket.
Closed SaaS
Quick to adopt — but your data, governance, and exit belong to someone else.
Framework wiring
Reusable parts — but you rebuild governance, identity, and audit per use case.
Oraclous — the fourth
Describe the goal in prose. The platform compiles it, governs it with ReBAC, and runs it across humans and Agents — open-source, portable, your keys.
One governed fabric
Humans and Agents, side by side on one task board.
Work hands off both ways — a person assigns an Agent; an Agent escalates to a person — and every step runs under one model of control. Human-in-the-loop isn't bolted on: it's a gate the runtime enforces.
Why it's different
A governed fabric, not another agent framework.
One workforce
Humans and AI Agents are symmetric Actors on one task board, under one governance model.
Describe the goal
Operators write goals in plain language; the platform compiles, governs, and runs them — no code per use case.
Your data, your keys
Self-host or cloud with identical guarantees. Every record is org-scoped; in cloud mode we can't decrypt your state.
Governance is the platform
ReBAC, policy sets, and provenance are substrate — enforced in code, not bolted on per project.
Bring any model
BYOM across Anthropic-native, OpenAI-compatible, and Gemini-compatible. Switch by config; the Harness never moves.
Open and honest
Open-source, ADR-documented, limits stated up front. Proof you can read — not metrics we invented.
The platform
One substrate. Every capability governed the same way.
Eight capabilities, one policy model — explore the platform, or jump straight in.
Harness model
Compose humans + Agents, tasks, policies, and triggers into one runnable unit.
Learn moreBring your own model
Run the model you choose; the LLM is a resource, not the Agent.
Learn moreKnowledge graph
Your org's provenance-tracked memory, retrieval bounded by ReBAC.
Learn moreReBAC governance
Access by relationship, enforced platform-wide; code wins over prose.
Learn moreHuman-in-the-loop
A first-class runtime primitive — waiting on a human is like waiting on a tool.
Learn moreExecution & scheduling
Durable runs, checkpoints, schedules, retries — no runaway loops.
Learn moreMCP & widgets
MCP server and client; embed Agents into your product, host-safe.
Learn morePortability
OHM exports your work — with the limits stated, so there are no surprises.
Learn moreEnterprise-grade by architecture
Your data stays yours — provably.
One layered platform on a substrate trust root: every record is organisation-scoped, credentials never leave the broker in plaintext, and in cloud mode we cannot decrypt your state. Governance (ReBAC) and audit are part of the substrate — not an add-on.
Proof, not promises
No invented metrics. No fake logos.
The proof is the architecture — open-source, decision-documented, with the trade-offs written down. You can read all of it before you commit a single workload.
- Open source
- ADR record
- Honest docs
- Per-service reference
- 5
- versioned policy sets
- 3
- model protocols (BYOM)
- 1
- organisation boundary
- 0
- token markup
Where do you start?
Pick the path that matches your work.
Questions
Frequently asked.
What is an AI agent orchestration platform?
An AI agent orchestration platform assigns, governs, and runs work across multiple AI Agents (and, in Oraclous, humans too) under one policy model — as substrate, rather than as plumbing each team wires by hand. Oraclous does this with Harnesses, ReBAC governance, and BYOM. More on the platform →
What is the difference between an AI agent framework and an AI agent platform?
A framework hands you parts and leaves governance, identity, credentials, audit, and metering for you to wire per use case. A platform makes those substrate — built once, enforced everywhere. Oraclous also inverts the unit of work: you describe a goal in prose rather than coding each agent. Why a platform, not a framework →
Can you self-host AI agents with Oraclous?
Yes. Oraclous is open source and platform-as-code, so you can run the whole platform on your own infrastructure with no vendor in the loop. The cloud-hosted mode carries identical data-sovereignty guarantees if you prefer operations off your plate — the choice of support model and isolation tier is yours. See the open-source story →
How do I avoid vendor lock-in with AI agents?
Two ways at once. BYOM keeps the model a resource you swap by config (Anthropic-native, OpenAI-compatible, Gemini-compatible), and OHM keeps your work portable — every export routes through the open manifest. The docs state plainly what portability does and doesn't carry, so there are no exit surprises. How portability works →
Is Oraclous open source and free?
The platform is open source and free to self-host — read the code, the ADRs, and the trade-offs before you commit. Cloud-hosted mode is a paid option with identical sovereignty guarantees and managed operations. Metering measures consumption neutrally; it does not set your prices. See pricing →
Form your second mind.
Open-source, sovereign, and governed by your own rules. Start with the architecture — read exactly how it works before you commit.