Gazebo

Gazebo is an identity and access management layer for AI agents at gazebohq.com ¹. Rather than pasting a master API key into every coding tool and automation, teams connect each service once to an encrypted vault, then issue every agent its own token scoped to only the services and HTTP methods it is permitted to use ^{1 2}. Every credential request is logged, and deleting an agent revokes its access immediately without rotating the underlying key ^{1 3}.

The framing is explicit: the access model most teams use for agents today is the one infrastructure had before IAM existed. Cloud providers solved it for humans and services years ago, and agents never got the same treatment ⁴.

Product

Gazebo works in three layers: the vault, access profiles, and agents ³.

1. Connect your services. API keys and OAuth tokens are added to the vault, encrypted at rest with AES-256-GCM and associated with a service. Plaintext credentials are never returned to a client after storage — the vault exists to issue them to permitted agents, not to display them back ³.
2. Create an access profile. A named permission set answering which services an agent can reach and what it can do with each. A stripe-reader profile with Stripe GET lets an agent make read-only Stripe calls and touch nothing else ³.
3. Create an agent and issue a bearer token. The token is an identity claim, not a credential — it carries no secret values ³.

At runtime the agent calls Gazebo via MCP (get_credential) or the REST API with its bearer token and the service it needs. Gazebo validates the token, checks the access profile against the requested service and method, logs the request, and returns either the credential or a structured denial. Credentials are fetched per request and never stored in agent config ³.

Workflows add an approval gate. Gazebo generates a step-by-step execution plan and shows it in full before anything runs; calls to production services execute only after an explicit approve, with no way for a workflow to bypass the gate ².

The site publishes setup guides for MCP-compatible clients including Cursor and Claude Code ⁵, and connects to services including Stripe, Cloudflare, Vercel, Resend and Google ¹. Separate integration pages cover existing secret stores such as 1Password, Doppler and HashiCorp Vault ⁶.

Pricing

All plans are free during the current beta ⁵:

Plan	Price	Key features
Free	$0	5 connected services, 2 access profiles, MCP/API access, 7-day audit retention, 5 workflow runs/mo
Pro	$9/mo	Unlimited services and profiles, 90-day audit retention, 50 workflow runs/mo, $0.10 per run over
Studio	$39/mo	Unlimited services and profiles, 1-year audit retention, 250 workflow runs/mo, 5 team members
Pro Lifetime	$99 once	Everything in Pro permanently, limited to 200 founding member spots

Team members are free during beta on every plan ⁵.

Security

Security is the product rather than a feature list, and the detail is published openly ²:

• Scoped agent identity. Every agent gets its own token, scoped to explicitly permitted services, with per-service HTTP method control. An operation outside the permitted methods is denied and raises an alert.
• Encrypted vault. Credentials are encrypted with AES-256-GCM before being written to the database, with the encryption key stored separately. Plaintext values are never logged, never returned to the client after submission, and never sent to the AI provider.
• Approval gate. Production API calls execute only after explicit human approval of the generated plan.
• Audit log. Every credential access is recorded with agent, service and timestamp. Every HTTP call an executed workflow makes is recorded with method, URL, response status and timestamp.
• Authentication. Passwords hashed with PBKDF2-SHA512 at 100,000 iterations with a random salt. Sessions use HTTP-only server-side cookies. Magic link tokens are single-use and expire after one hour.
• Programmatic access. REST API and MCP endpoints require a bearer token. Full-account tokens grant broad access; agent tokens are scoped to permitted services. Both are revocable from account settings.

Who it’s for

• Solo developers running AI coding tools who currently keep production keys in a .env file shared across every agent they run, and want per-agent scoping without standing up Vault.
• Small teams shipping with agents who need one audit trail across services rather than reconciling Stripe restricted keys, GitHub token scopes and OpenAI project keys separately.
• Anyone granting an agent write access to production who wants an approval step between the model’s plan and a live API call.

How it compares

Conventional secret managers store credentials and hand them to whatever process asks with the right token. Gazebo’s distinction is that the consumer is modelled as an agent with an identity, a permission profile and an audit history, and that the permission check happens per request at the point of retrieval rather than at deploy time ³.

It is positioned alongside existing vaults rather than purely against them: 1Password, Doppler, Infisical, HashiCorp Vault and AWS Secrets Manager each get a dedicated integration page ⁶.

Alternatives

• Choose HashiCorp Vault if you need a mature, self-hostable secrets platform with dynamic secrets and broad enterprise integration, and have the operational capacity to run it.
• Choose Doppler or Infisical if your main problem is syncing environment configuration across developers and deployment targets rather than governing what an autonomous agent may call.
• Choose 1Password if the primary need is human credential storage with team sharing, and agent access is incidental.
• Choose Gazebo if the thing holding your keys is an AI agent, and you want per-agent scoping, method-level limits, an approval gate and a single audit trail of what each agent actually touched.

Verified performance

Lighthouse audit of gazebohq.com (August 2026) ⁷:

Metric	Score
Performance	100
SEO	100
Accessibility	100
Best Practices	77
LCP	0.7 s
FCP	0.6 s
CLS	0.002
TBT	0 ms

Company

Gazebo was founded by Ash Conway, listed as Founder and CEO ⁴. He previously founded Bugwolf in 2013 after a decade in enterprise software and hardware at Cisco, SAP and Dell, and bootstrapped it to multimillion-dollar revenue without outside funding ⁴.

The stated origin is that per-service access controls stopped composing once teams ran multiple agents: no unified view of which agent held which keys, no cross-service audit trail, and no way to cut one agent off without touching every service separately. Gazebo is built as the missing layer between agents and the services they call ⁴.

The product is in beta, with all plans free during the beta period ⁵. A public status page is maintained at gazebo.instatus.com ⁴.

External links

• gazebohq.com (official site)
• About
• Pricing
• Security
• How Gazebo works
• Integrations
• Status page

Footnotes

1. Gazebo homepage. Retrieved August 1, 2026. https://gazebohq.com ↩ ↩² ↩³ ↩⁴

2. Gazebo security page. Retrieved August 1, 2026. https://gazebohq.com/security ↩ ↩² ↩³

3. How Gazebo works, Gazebo docs. Retrieved August 1, 2026. https://gazebohq.com/docs/how-it-works ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

4. Gazebo about page. Retrieved August 1, 2026. https://gazebohq.com/about ↩ ↩² ↩³ ↩⁴ ↩⁵

5. Gazebo pricing page. Retrieved August 1, 2026. https://gazebohq.com/pricing ↩ ↩² ↩³ ↩⁴

6. Gazebo integrations index. Retrieved August 1, 2026. https://gazebohq.com/integrations ↩ ↩²

7. Lighthouse audit of gazebohq.com, run August 1, 2026. ↩