Alma
Self-hosted AI agent governance with an access map, Cedar policy enforcement and signed audit evidence.
The access map is the reason to look here: permitted-versus-observed drift on a per-resource basis is something a static config review simply cannot produce. Add four deny-closed enforcement points and a signed, hash-chained ledger and you have governance that survives an auditor asking how you know. The honest caveats are operational: you run the control plane yourself, and coverage depends entirely on which sources you connect and what they record.
Verified 3d ago · liveness 79/100 · cite: rightaichoice.com/tools/alma
- Platform and DevOps teams self-hosting their own AI agent infrastructure
- Security teams that need deny-closed enforcement plus auditor-ready evidence
- CISOs who require vendor-neutral governance and data residency control
- Organizations with air-gap or strict data-residency requirements
- Buyers who want a fully managed cloud governance service with no control plane to run
- Organizations that cannot adopt AGPL-3.0 and won't buy the commercial Business licence
- Estate owners needing multiple active production instances outside an Enterprise order form
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Olivares AI if your organization cannot adopt AGPL-3.0 and won't buy the commercial licence, or if you want a hosted vendor to operate the governance control plane for you.
Business is US$129/month billed monthly or US$1,290/year billed annually — the yearly price is two months free, so a team that starts monthly and switches later pays more for the first year.
Community is free under AGPL-3.0 with unlimited users — the whole platform, not a cut-down edition — which makes it cheaper than any per-seat governance tool for large teams. For commercial deployment, Business at US$129/month or US$1,290/year per subscription is priced per deployment rather than per user, so it stays flat as headcount grows. Enterprise is custom annual order-form pricing for several legal entities, OEM or redistribution rights. Hosted governance suites run by cloud vendors
In short
Alma — Self-hosted AI agent governance with an access map, Cedar policy enforcement and signed audit evidence. Best for Platform and DevOps teams self-hosting their own AI agent infrastructure, Security teams that need deny-closed enforcement plus auditor-ready evidence, CISOs who require vendor-neutral governance and data residency control. Free to start; paid plans from $129/mo.
What's new in Alma
Checked yesterdayAcross the latest 2 updates: 1 feature update and 1 changelog entry.
Run Claude Code, Codex and Grok sessions on your own server from the browser
Olivares documents starting, following, approving and stopping Claude Code, Codex and Grok sessions on self-hosted infrastructure via web console or CLI, using existing accounts.
Olivares AI 0.2: Claude Code session activity, slash commands, skills lifecycle
Version 0.2 adds a Claude Code session Activity tab, slash-command menu, and full skills-pack lifecycle (update, diff, pin, enable/disable) across CLI, console and API. Access map, identity inventory and workspace wiki move to Community.
What people actually say about Alma — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
69 mentions across 4 sources (Hacker News, Product Hunt, App Store, Lemmy) · researched Jul 3, 2026.
Average across the 4 sources that answered — each source counts once, not each post.
- +Deny-closed default with Cedar-based RBAC/ABAC is security-best-practice approach.
- +Passive discovery of agents and MCP servers without proxies or agents.
- +Tamper-evident audit ledger with cryptographic chaining ensures compliance readiness.
- +Emergency kill switch for instant session revocation adds critical safety.
- +FinOps module for per-agent budgets enables cost allocation and control.
- −No community feedback available to validate actual usability or stability.
- −Feature claims are extensive but unverified in real-world deployments.
- −AGPL license with 3-user cap may complicate evaluation and adoption.
- −Documentation and onboarding experience cannot be assessed from data.
- −Self-hosted setup may require significant DevOps effort for non-experts.
- • Self-hosting infrastructure and maintenance costs not included.
- • Commercial license pricing undisclosed; may require sales contact.
Viability Score
How well maintained and how widely used is Alma? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Passive discovery of agents, sessions, MCP servers and models from connected sources
- Per-resource read vs read/write access map built from connected evidence
- Permitted-vs-observed least-privilege drift surfaced as classified findings
- Policy as code in Cedar with allow, deny and on-violation actions
- Deny-closed enforcement at the Claude Code PreToolUse/PostToolUse hook
- Inline inference proxy enforcement on agent model calls
- MCP tools/call gate secured with the OAuth 2.1 flow MCP mandates
- A2A delegation gate for agent-to-agent handoffs
- Official Codex CLI and Grok CLI as session drivers (v26.9.0)
- Provider profiles giving each configured provider instance a durable identity
- Interrupt a live session turn from the console with the operator recorded
- Append-only hash-chained audit ledger with per-event Ed25519 signatures and signed checkpoints
- Audit export to CEF, LEEF, syslog, OTLP and OSCAL
- Push connectors to Splunk HEC, Microsoft Sentinel, Datadog, New Relic and HMAC webhooks
- FinOps budgets that deny or throttle spend, attributed per agent, model and team
About Alma
Olivares AI (Alma) is a self-hosted, open-core control plane for teams running AI coding agents on their own infrastructure. Instead of asking you to route everything through a hosted service, it installs as a binary or container using SQLite or PostgreSQL, connects to observation sources, and reports back what your agents, sessions, MCP servers and models actually are. The core question it answers is a blunt one: what can these agents reach, and can you prove what you stopped? The differentiator is the access map. Olivares builds a per-resource read versus read/write picture from the sources you connect, then diffs permitted access against observed access and surfaces least-privilege drift as classified findings. Policy is written as code in Cedar with allow, deny and on-violation actions, enforced at four deny-closed points rather than reported after the fact: the Claude Code PreToolUse/PostToolUse hook, an inline inference proxy, an MCP tools/call gate, and an A2A delegation gate for agent-to-agent handoffs. As of v26.9.0 the official Codex CLI and Grok CLI join Claude Code as session drivers, provider profiles give each configured provider a durable identity, and an operator can interrupt a live session turn from the console. Evidence is the second half of the pitch. Recorded events land in an append-only, hash-chained ledger with per-event Ed25519 signatures and signed checkpoints, then export to CEF, LEEF, syslog, OTLP and OSCAL with push connectors to Splunk HEC, Microsoft Sentinel, Datadog and New Relic. Cost governance attributes model spend to the agent, model and team that caused it, and budgets can deny or throttle rather than simply report. Who it suits: platform, security and compliance teams in regulated or air-gapped networks. Community is the complete product under AGPL-3.0 with unlimited users and one active identity provider; Business adds the commercial licence and the signed release channel at US$129/month or US$1,290/year per subscription,
Behind the Verdict
Most agent governance products want your traffic. Olivares AI starts from the opposite premise — install a binary on infrastructure you control, connect sources you already have, and read the inventory. We'd reach for this when the estate is regulated, air-gapped, or simply not something you can hand to a third-party control plane. Where it earns its keep is the access map. A permitted-versus-observed diff, expressed per resource as read or read/write and surfaced as a classified finding with a confidence level, is the artifact that turns "we think our agents are scoped" into something reviewable. The four enforcement points matter because they are deny-closed: the Claude Code hook, the inline inference proxy, the MCP tools/call gate, and the A2A delegation gate. Pair that with the append-only ledger, per-event Ed25519 signatures and signed checkpoints, and the export story to CEF, LEEF, syslog, OTLP and OCSF becomes genuinely useful rather than decorative. Cost governance is the sleeper feature. Budgets that deny or throttle, attributed to the agent, model and team that generated the spend, are worth more than a dashboard that tells you next month what happened last month. The practical caveats are real. Inventory is passive, which means it is only as complete as your connected sources; the vendor says this plainly and so should you when you scope a rollout. Enforcement applies at Olivares' enforcement points, not to every operation of an external tool. And you are operating a control plane: TLS, identity, OIDC or SAML config, upgrades, storage. If nobody on the team wants that job, this isn't the product. On licensing, know your position before you start evaluating. Community is the complete platform under AGPL-3.0 with unlimited users and one active identity
Researching Alma? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Alma actually fits — and what changes day-one when you adopt it.
Stand up the single Go binary against the built-in demo estate, wire the Claude Code hook and a pgaudit source, then walk the read/write access map until the unreviewed write to prod-postgres from a data-export job surfaces as a drift finding.
Outcome: A populated access graph in about five minutes on demo data, then a real estate inventory with least-privilege drift visible as a classified finding rather than a config-review guess.
Write a Cedar policy pinning the export job to read-only on the production database with on_violation set to block and alert, let it enforce at the hook and MCP gate, then export the hash-chained ledger to Splunk HEC and OSCAL.
Outcome: The write is denied at the enforcement point instead of logged after the fact, and the auditor receives Ed25519-signed evidence an external verifier can confirm offline.
Configure budgets per team against provider cost streams, attribute the last 30 days of spend to the agents and models that caused it, and set the budget gate to throttle rather than only report when a team crosses its ceiling.
Outcome: Spend stops being an estimate: budgets deny or throttle at the cap, and the platform team can see that platform was at 91% and finance at 51% of their 30-day budgets before the next invoice.
Use Cases
- Inventory every AI agent, session, model and MCP server on your estate without deploying a proxy on each host.
- See the diff between what an agent is permitted to touch and what it is actually reading and writing.
- Pin an agent to read-only on a production database in Cedar and block writes at the enforcement point.
- Prove to an auditor who touched what, with a hash-chained, Ed25519-signed ledger exported as evidence.
- Cap AI spend per agent, model and team with budgets that deny or throttle when a cap is hit.
- Govern Claude Code tool-calls deny-closed while keeping governance data on-prem.
- Handle retention, erasure and legal hold with a WORM archive in your own bucket.
- Run an air-gapped deployment where no governance data leaves your infrastructure and licence validation is offline.
Models Under the Hood
as of 2026-10-17
Limitations
- Self-hosted, open-core governance plane: you run and operate the infrastructure yourself; Community (AGPL-3.0) is free with unlimited users, one active identity provider, and public-repo support with no SLA, while Business is a commercial licence at US$99.00/month or US$990.00/year per subscription, not per user.
- The vendor states visibility and enforcement depend on the integration and the controls you configure, and describes compliance evidence as control status and drafts of EU AI Act, ISO/IEC 42001, DORA and NIS 2 reporting rather than a certification.
- The platform governs the AI you already run with your existing tools — multi-provider by design — rather than supplying models itself; air-gapped install is Enterprise.
as of 2026-10-09
Verification history
We have re-verified Alma 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Alma tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Community (AGPL-3.0)
Free
Ideal for
Teams that can adopt AGPL-3.0 and are willing to run the control plane themselves — from one engineer to a whole company, since there is no user cap or licence key.
What this tier adds
Free entry point: the complete self-hosted platform with passive discovery, access map, drift findings and one active identity provider, but public repository support and no SLA.
Business (Monthly)
US$129/month
Business (Yearly)
US$1,290/year
Enterprise
Custom
Ideal for
Organizations needing negotiated terms, several legal entities, more than two productions, more than five identity providers, OEM or MSP redistribution, or air-gapped and LTS deployments.
What this tier adds
Adds custom credential exchange with scoped credential minting, scoped dedicated engineering for policy authoring and connectors, LTS with air-gap and OTA mirror, OEM and redistribution rights, capped indemnity and DPA, and best-effort email first-response targets under an
Where the pricing makes sense
The company stage and team size where Alma's pricing actually pencils out — and where peers do it cheaper.
Community is free under AGPL-3.0 with unlimited users — the whole platform, not a cut-down edition — which makes it cheaper than any per-seat governance tool for large teams. For commercial deployment, Business at US$129/month or US$1,290/year per subscription is priced per deployment rather than per user, so it stays flat as headcount grows. Enterprise is custom annual order-form pricing for several legal entities, OEM or redistribution rights. Hosted governance suites run by cloud vendors
Setup time & first value
How long it actually takes to get something useful out of Alma — broken out by persona, not the marketing-page minute.
From zero to a populated read/write access graph is about five minutes against the built-in demo estate — one static binary, no external services. A production install is longer: the docs cover the single binary install with TLS on by default, /readyz and /livez probes, SQLite or least-privilege Postgres, then identity configuration and source wiring. Budget hours for identity, sources and the
Switching to or from Alma
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From spreadsheet or manual config review: connect a read-only observation source and let passive discovery replace the manual inventory of agents, sessions, MCP servers and models.
- →From Claude Code hooks configured by hand: move tool-call gating into Cedar policies at the PreToolUse/PostToolUse enforcement point and keep the same hook-level depth.
- →From a SIEM-only audit trail: keep Splunk HEC, Microsoft Sentinel, Datadog or New Relic as the destination while the hash-chained ledger provides the signed evidence layer the SIEM does not.
- →From pgAudit log review alone: stream pgAudit events into the access map so read vs read/write edges get attribution and permitted-vs-observed drift is computed for you.
- →From static permission audits: import the grants you already documented and let the observed side of the map tell you where reality diverges.
- ↗To a hosted governance suite: export the audit record to OCSF or OSCAL first, since a managed vendor will not carry your self-hosted ledger across.
- ↗To Splunk or Microsoft Sentinel as the governance record: keep the push connector running long enough to confirm the exported event stream matches the hash-chained ledger before you decommission the control plane.
- ↗To manual Claude Code hooks: Cedar policies translate into hook configuration, but you lose the A2A delegation gate, the MCP tools/call gate and the passive inventory.
- ↗To no governance layer: run the kill-switch drill and export sealed evidence one last time so you have a defensible record of the period the platform was in force.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Alma”, and we withheld 6: 6 could not be judged, because “Alma” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Alma.
Official links
Featured Head-to-Head Comparisons
Alma vs Temporal Ai
Choose Temporal if your core pain point is building reliable, failure-tolerant AI agents and workflows that survive crashes and retries. Choose Alma if you already have AI agents running and need visibility, governance, least-privilege enforcement, and an audit trail—Temporal does not provide governance controls. They are complementary: Temporal orchestrates; Alma governs.
Alma vs Audioeye
Alma and AudioEye serve completely different markets—Alma is an open-core self-hosted platform for governing AI agent security and compliance, while AudioEye is a paid SaaS for web accessibility compliance. Your choice depends solely on whether you need AI governance or accessibility remediation; there's no overlap.
Alma vs Push Security
Choose Push Security if your primary threat is browser-based attacks (AiTM, ClickFix, session hijacking) and you need to control AI tool data leakage across any browser without migration. Choose Alma if you're a DevOps or security team governing AI agents (Claude Code, MCP servers) in a self-hosted environment and require least-privilege enforcement with an immutable audit trail. They are complementary: Push protects the user endpoint; Alma protects the agent infrastructure.
Popular in AI Governance & Guardrails
Mindgard
Mindgard automates AI red teaming — agent-native reconnaissance, exploit validation, and continuous AI security testing for production AI
Poolside AI
Poolside builds open-weight agentic coding models — Laguna XS 2.1 (33B) and Laguna S 2.1 (118B) — for code that stays inside your security
Olas Network
Co-own autonomous AI agents on-chain: stake OLAS, run agents that trade, hire each other and settle payments.
Frequently Asked Questions
Best-of guides
Used Alma? Help shape our editorial sentiment research.