mTLS between components
Mutual TLS on every internal hop; no plaintext control traffic.
● enforcedSelf-hosted · Vendor-neutral · Open-core
Integrate, manage and secure the AI you actually run — one ground truth.
Context, resources, sessions and identities in one self-hosted plane: Claude Code at the deepest level, Codex and Grok Build alongside — and open to any other model or provider you run.
Your agents get what they need to do real work; you keep the granular permissions, policy, budgets and audit evidence to run all of it. It complements your agents; it does not compete.
The product tour
Explore the estate, area by area
Every surface, with real console screenshots and an honest account of what is live today.
Access map
Read/write access and least-privilege drift — the differentiator.
Live operations & sessions
Inspect observed session actions, models, tokens, costs, and event history, with governed work items and handoffs. Goals depend on the reporting source; computed task progress and the complete operating workflow are not yet verified.
Kill switch
One-click estate stop, dual-control recovery.
Compliance evidence
Control status and sealed evidence — not a certification.
Cost & FinOps
See, attribute and govern model spend.
Real screenshot
See the real product
17 genuine views of the Olivares console, captured from the product with example data.
- 30
- wired modules · one self-hosted release
- 100%
- self-hosted — governance data stays under your control
- 0
- mandatory calls home · runs air-gapped
- R / RW
- per-resource access · permitted vs observed
Discover · 01 / 05
Every agent in your estate, discovered passively
No proxy, no agents to babysit. Olivares watches your infrastructure and inventories every agent, session, model and MCP — with the surfaces each one reads and writes.
data-export-job holds an unreviewed write to prod-postgres — least-privilege drift, surfaced the moment it appears.
Govern · 02 / 05
One rule, before and after
Write policy as code and watch it land on the map. The unreviewed write to your production database becomes a denied, least-privilege path.
policy · prod-db-writes
01policy "data-export-readonly" {02 agent = "data-export-job"03 resource = "prod-postgres"04 allow = ["read"]05 deny = ["write"]06 on_violation = "block + alert"07}
Pin the export job to read-only on the production database; block writes.
enforcement● enforcing
Cost · 03 / 05
Spend, attributed — not estimated
Every dollar tied to the agent, model and team that caused it — with budgets and trends your finance team can act on, not guess at.
Spend by agent last 30 days
| Agent | Model | 30d | Trend | % total |
|---|---|---|---|---|
| claude-deploy-bot | claude-opus-4-8 | $1,860 | 22.1% | |
| data-export-job | magistral-small | $1,240 | 14.8% | |
| support-rag-agent | gpt-5.5 | $980 | 11.7% | |
| infra-copilot | claude-opus-4-8 | $880 | 10.5% | |
| billing-reconciler | claude-sonnet-4-6 | $720 | 8.6% | |
| qa-runner | gpt-5.4-mini | $180 | 2.1% | |
| +208 more agents | $2,540 | 30% | ||
Audit · 04 / 05
An append-only record of who touched what
Every access in a tamper-evident, hash-chained ledger — filter it, replay it, and export it as evidence your auditors accept.
Security & Compliance · 05 / 05
Verifiable controls, mappable frameworks
mTLS, an append-only ledger and policy enforced at access time — mapped to the frameworks your auditors ask about.
Append-only ledger
Every view and change hash-chained and tamper-evident.
● activeGuardrails
Policy enforced at access time — blocked, not just logged.
● enforcedLeast-privilege collector
Passive discovery from a minimally-privileged, read-only collector.
● active Framework coverage mapped partial in progress
| Framework | Access control | Audit trail | Risk mgmt | Data governance | Coverage |
|---|---|---|---|---|---|
| EU AI Act | mapped | mapped | mapped | mapped | 96.9% |
| NIST AI RMF | mapped | mapped | mapped | partial | 98.6% |
| ISO 42001 | mapped | mapped | in progress | mapped | 97.4% |
| SOC 2 | mapped | mapped | partial | mapped | 98.4% |
coverage 98% Olivares maps your controls to each framework; coverage shown is self-assessed and illustrative — not a certification. It gives you the evidence to support your certification path.
How it works
Running in your own infrastructure
- 01
Deploy
A single container or binary on your own infrastructure. No account, no cloud dependency, no mandatory calls home.
- 02
Discover
Olivares AI passively finds every agent, session, MCP and model already running — no mandatory proxy.
- 03
Map
It builds the access graph: what each agent can reach and what it actually touches, read versus read/write.
- 04
Govern
Set policy, watch for drift and export audit evidence. Visibility becomes control as your estate grows.
vendor-neutral · runs across Claude OpenAI Gemini Llama Ollama / vLLM Postgres S3 Vault Kafka Kubernetes AWS / Cloudflare
Open source
Your infrastructure. Your data. Your ground truth.
Olivares AI is open-core under AGPL-3.0. The complete product is in the public repository, tagged and signed since v26.8.0 — to run on your own infrastructure, inspect every line, and keep your data where it belongs. A commercial license and a dedicated enterprise tier are available when your organization needs them.
-
AGPL-3.0
The complete product, self-hosted, free — no feature gates on the core.
-
Vendor-neutral
Not tied to any single AI vendor, identity provider or cloud.
-
Air-gapped friendly
Olivares makes no mandatory calls home and runs in isolated, regulated networks. Fetching and updating the commercial add-ons needs an active subscription.
Support the project.
Sponsorship sustains the development, integration and updates of Olivares AI; it is not a support contract and buys no priority.
See what your agents can reach
Deploy Olivares AI on your own infrastructure and get the access map your platform and security teams have been asking for.