Metorial vs Archestra

Archestra and Metorial both put a governed layer between AI clients and the tools your company uses.

Where Archestra wins: LLM proxy with hard spend limits per organization, team, user, agent, virtual key or environment; requests are blocked with HTTP 402 once a limit is hit.

Metorial is the stronger fit if you want Git-synced skills, ChatGPT support and an EU data region.

Every Archestra fact below links to Archestra's own page. Anything they don't publish is marked "Not published", not "No".

Where Archestra wins

  • LLM proxy with hard spend limits per organization, team, user, agent, virtual key or environment; requests are blocked with HTTP 402 once a limit is hit.source
  • Enterprise features (SSO, RBAC, granular access control) are free in production for companies under 30 users.source
  • Session-level deterministic guardrails: tool results add restrictions to the session and later tool calls must satisfy them, enforced outside the agent loop.source

Key differences

Each cell links to the page it comes from.

FeatureMetorialArchestra
Source licenseFSL-1.1-ALv2 engine (Apache 2.0 after two years); Apache-2.0 integration catalogsourceThe engine (metorial-platform) is source-available under FSL and converts to Apache 2.0 two years after each release. The integration catalog (metorial/metorial) is Apache-2.0.AGPL-3.0 (enterprise features under separate LICENSE_ENTERPRISE)source
Self-hostYessourceYessource
Integration catalog1,000+source900+source
One URL for all toolsYessourceYessource
Per-user OAuthYessourceYessourceREADME: OAuth + On-Behalf-Of so tools run as the user.
SSO (SAML or OIDC)YessourceYessourceOkta, Entra ID, Google, GitHub, GitLab, generic OIDC/SAML 2.0; enterprise license (free under 30 users).
SCIM provisioningYessourceNot publishedTeam sync from directory groups at SSO login; SCIM not mentioned.
Role or group-based tool accessYessourceYessourceGranular access control is an Enterprise feature.
Audit log per personYessourceYessourceMCP Gateway log records tool, gateway, account, arguments and result per call.
Create and edit skillsYessourceYessource
Git sync for skillsYessourcePartialsourceImport documented; ongoing sync not stated.
Own chat app requiredNosourceNosourceShips its own chat & projects UI, but external clients supported.
Prompt-injection detectionYessourcePartialsourceMitigation by deterministic session policy rather than detection; Guardrails in Beta and AI Security tier.
EU data regionYessourceNot publishedNo managed cloud offering documented; self-host only.
SOC 2 Type IIYessourceNot published
LLM gateway or model routingNoYessourceLLM Proxy with providers, model router, virtual keys.
Starting price$0 (Dev); Scale from $250/monthsourceFree (AGPL; enterprise features free under 30 users); Enterprise contact salessource

Which AI clients each works in

FeatureMetorialArchestra
Claude (web and desktop)YessourceYessourceClaude Desktop documented; claude.ai web not explicitly.
Claude CodeYessourceYessource
ChatGPTYessourceNot published
CursorYessourceYessource
CodexYessourceYessource
GitHub CopilotYessourcePartialsourceCopilot CLI named; Copilot in IDE not named.
Custom agents (SDK or API)YessourceYessourceDocs section 'AI SDKs and Frameworks'.

Full checklist

Every row we track, with what it means. "Not published" means the vendor's own pages don't say.

Connect

FeatureMetorialArchestra
Integration catalogHow many ready-made integrations or MCP servers the vendor publishes.1,000+source900+source
Verified or official serversThe catalog includes servers the vendor tests or maintains, not only community listings.YessourcePartialsourceCatalog servers are 'evaluated'; no formal verification program described.
Bring your own MCP serverYou can add an MCP server you built or host yourself.YessourceYessource
Remote (HTTP) MCP serversConnects to MCP servers hosted elsewhere over HTTP.YessourceYessource
Local, stdio or Docker serversRuns MCP servers that ship as local processes or containers.YessourceYessourceSelf-hosted servers run as Kubernetes workloads (embedded cluster in Docker quickstart).
One URL for all toolsA user adds one endpoint to their AI client and reaches every approved tool through it.YessourceYessource
Version pinningYou can lock an integration to a known version.YessourceNot published
Schema change monitoringChanges to tool definitions are detected and flagged.YessourceNot published

AI clients

FeatureMetorialArchestra
Claude (web and desktop)Documented to work in the Claude apps.YessourceYessourceClaude Desktop documented; claude.ai web not explicitly.
Claude CodeDocumented to work in Claude Code.YessourceYessource
ChatGPTDocumented to work in ChatGPT.YessourceNot published
CursorDocumented to work in Cursor.YessourceYessource
CodexDocumented to work in Codex.YessourceYessource
GitHub CopilotDocumented to work in GitHub Copilot.YessourcePartialsourceCopilot CLI named; Copilot in IDE not named.
Custom agents (SDK or API)Your own agents can use the tools through an SDK or API.YessourceYessourceDocs section 'AI SDKs and Frameworks'.
Own chat app requiredYes means people must work inside the vendor’s own app rather than the AI client they already use.NosourceNosourceShips its own chat & projects UI, but external clients supported.

Identity and access

FeatureMetorialArchestra
Per-user OAuthTool calls run with each person’s own account, not a shared key.YessourceYessourceREADME: OAuth + On-Behalf-Of so tools run as the user.
Managed OAuth appsThe vendor provides OAuth apps so you don’t register your own.YessourceNot published
Bring your own OAuth appYou can use OAuth apps registered under your own company.YessourceNot published
SSO (SAML or OIDC)People sign in through your identity provider.YessourceYessourceOkta, Entra ID, Google, GitHub, GitLab, generic OIDC/SAML 2.0; enterprise license (free under 30 users).
SCIM provisioningUsers and groups sync automatically from your identity provider.YessourceNot publishedTeam sync from directory groups at SSO login; SCIM not mentioned.
Role or group-based tool accessAccess to tools follows roles or groups.YessourceYessourceGranular access control is an Enterprise feature.
Tool-level allow and denyIndividual tools can be allowed or blocked, not only whole servers.YessourcePartialsourcePer-gateway tool selection (Manual mode); no explicit per-tool allow/deny rules documented.
Read-only modesAn integration can be limited to read-only tools.YessourceNot published'Can read' permission level applies to platform resources, not tool calls.

Visibility and security

FeatureMetorialArchestra
Audit log per personEvery call is recorded and tied to the person or agent who made it.YessourceYessourceMCP Gateway log records tool, gateway, account, arguments and result per call.
Call tracingTool arguments and results are recorded for debugging and review.YessourceYessource
Log export or SIEMLogs can be exported or streamed to your security tools.YessourcePartialsourceMetrics and traces export (homepage: OpenTelemetry and Prometheus); no named SIEM integration.
Data redaction or DLPSensitive data in tool traffic can be detected, masked or blocked.YessourcePartialsourceMetadata-only logging mode; guardrails can redact secrets; no PII/DLP feature documented.
Prompt-injection detectionTool traffic is inspected for prompt-injection attacks.YessourcePartialsourceMitigation by deterministic session policy rather than detection; Guardrails in Beta and AI Security tier.
Tool-poisoning protectionMalicious or changed tool descriptions are detected.YesThrough Protoguard.Not published

Skills

FeatureMetorialArchestra
Create and edit skillsTeams can write and edit shared skills in the product.YessourceYessource
Git sync for skillsSkills sync to or from a Git repository.YessourcePartialsourceImport documented; ongoing sync not stated.
Skills delivered by roleSkills reach people based on their role or group.YessourceNot publishedSkills scoped by user, agent and environment; roles/teams not mentioned on skills page.
Skills over MCPSkills are served to AI clients through MCP.Not publishedPartialsourceShared marketplace for coding agents; transport not explicitly MCP.

Deployment

FeatureMetorialArchestra
Source licenseThe license the product’s code is published under, if any.FSL-1.1-ALv2 engine (Apache 2.0 after two years); Apache-2.0 integration catalogsourceThe engine (metorial-platform) is source-available under FSL and converts to Apache 2.0 two years after each release. The integration catalog (metorial/metorial) is Apache-2.0.AGPL-3.0 (enterprise features under separate LICENSE_ENTERPRISE)source
Self-hostYou can run it on your own infrastructure.YessourceYessource
VPC or on-premCan be deployed in your own cloud account or data center.YessourceYessource
Air-gappedCan run with no internet connection.YesNot published
EU data regionData can be hosted in the EU.YessourceNot publishedNo managed cloud offering documented; self-host only.

Compliance

FeatureMetorialArchestra
SOC 2 Type IIHolds a SOC 2 Type II report.YessourceNot published
ISO 27001Holds ISO/IEC 27001 certification.Not publishedNot published
HIPAASupports HIPAA workloads, usually with a BAA.YesSigns BAAs.Not published
GDPR and DPAStates GDPR compliance and offers a data processing agreement.YessourceA data processing agreement is published at /legal/data-processing-agreement.Not published

Pricing

FeatureMetorialArchestra
Public pricingPrices are published on the vendor’s site.YessourceNosourceFree terms public; paid license price not listed.
Free tierA free plan or free edition exists.YessourceYessourceApplies to companies with fewer than 30 Archestra users.
Starting priceThe lowest published paid price, as listed.$0 (Dev); Scale from $250/monthsourceFree (AGPL; enterprise features free under 30 users); Enterprise contact salessource
Pricing modelHow the vendor charges: per seat, usage, platform fee, or a mix.Flat monthly plans with paid add-ons (extra seats, tool calls)sourceOpen core: AGPL base + Enterprise license required at 30+ userssource

Model layer

FeatureMetorialArchestra
LLM gateway or model routingRoutes model traffic, not only tool traffic.NoYessourceLLM Proxy with providers, model router, virtual keys.
Cost tracking per personShows AI spend per person or team.NoYessourceBudgets/limits per org, team, user, agent, virtual key, environment.

Which should you pick?

SituationPickWhy
Rolling AI out to every employee in the clients they already useEitherBoth connect to existing AI clients. Compare the client list below.
You must self-hostEitherBoth document a self-hosted option.
You need an LLM gateway or model routing todayArchestraArchestra documents an LLM gateway. Metorial doesn't ship one yet.
Teams should share skills, not only toolsEitherBoth document shared skills; compare the Skills rows below.
You are building integrations into your own productEitherBoth offer an SDK or API for your own agents.

Pricing

Published prices only. Nothing is estimated.

Metorial

  • Metorial Dev: $0 monthsource
  • Metorial Scale: $250 monthsource
  • Metorial for Enterprise: Contact salessource

Archestra

  • Open source (AGPL-3.0): Free self-hostedsource
  • Enterprise (small-team exception): Free companies with fewer than 30 userssource
  • Enterprise: Contact sales license (30+ users)source
  • Platform + AI Security: Contact sales licensesource

What each Metorial plan includes

From the Metorial pricing page.

IncludedMetorial DevMetorial ScaleMetorial for Enterprise
Tool calls per month500,0002,500,000Custom
Callbacks10,000100,000Custom
Workforce seats220Custom
Provider integrations10UnlimitedUnlimited
Log retention30 days90 daysCustom
Production accessNot includedIncludedIncluded

On every plan

  • Slack and email support
  • Auth Connections: Unlimited
  • Sessions: Unlimited
  • Concurrent Connections: Unlimited
  • Developer API Access
  • Custom Integrations
  • Remote MCP Servers
  • Managed MCP Servers

Archestra limits

As published by the vendor.

  • Free enterprise features: Fewer than 30 users company-wide; exception ceases at 30+source
  • Spend limit enforcement: Blocks requests once spend reaches limit (HTTP 402 token_cost_limit_exceeded)source

Moving from Archestra to Metorial

  • Remote MCP servers you use through Archestra can be added to Metorial as they are.
  • OAuth apps registered under your company can be reused, so people sign in the way they do today.
  • Sign-in moves to your identity provider through SAML; groups decide who gets which tools.
  • Skills, policies and audit history in Archestra don't transfer automatically. Talk to us about moving them.

Book a demo to plan the move with us.

Frequently asked questions

Common questions about Metorial and Archestra.

  1. Is Archestra open source?
    Archestra lists: AGPL-3.0 (enterprise features under separate LICENSE_ENTERPRISE).
  2. Yes, Archestra documents a self-hosted option.
  3. Archestra's own pages don't document ChatGPT support.
  4. Metorial's Dev plan is free and Scale starts at $250 a month. Archestra doesn't publish prices, so a direct comparison needs a quote.
  5. Yes. Metorial is SOC 2 Type 2 and GDPR compliant, with EU and US regions.
  6. Metorial's engine is published under FSL-1.1-ALv2 engine (Apache 2.0 after two years); Apache-2.0 integration catalog. You can self-host it or use the hosted platform.

Checked October 17, 2026. 34 Archestra facts are sourced, each linked in its row. Spotted something wrong? Tell us and we will fix it. All product names are trademarks of their owners.

Try Metorial next to Archestra

Start free, self-host from GitHub, or talk to us about your rollout.