Clawvisor vs Dome Systems (2026)

Fourteen service adapters let Clawvisor execute authenticated actions in outside systems on the agent's behalf. They cover Gmail, Google Calendar, Drive and Contacts. For messaging there are iMessage, Slack and Twilio, and for work tools GitHub, Notion, Linear and Granola, plus Stripe, Dropbox and Perplexity.

Each adapter takes the agent's requested action, attaches the real credential from the vault inside the gateway and returns the result. Customers can write their own adapters against the source for internal APIs. Adapters both read and write, with credentials scoped per task and revoked when it ends. SendGrid, Jira, Salesforce and Airtable are coming soon.

SourceClawvisor, clawvisor.com/integrations and /faqread 2026-09-25

The Tool Gateway is how agents act on outside systems. Admins attach MCP servers, REST APIs and internal services with their upstream authentication, OAuth included, and tools appear in a catalog for each agent, filtered by policy. Each call goes out through Dome with the credential injected at egress, so the agent never holds it, and customers can add tools of their own.

SourceDome Systems, domesystems.ai llms.txt (platform/gateway) and docs.domesystems.ai llms.txt (connect/resources/tools)read 2026-09-25

The product authorizes an agent's calls and does not run its work. A task is the authorized unit of work, made up of a purpose, the allowed tool calls and an approver. Each task moves through approval, scope expansion and expiry. The steps, their order and any handoffs between agents stay with the agent itself.

SourceClawvisor, clawvisor.com/faq and the README at github.com/clawvisor/clawvisorread 2026-09-25

The agent's own code owns the loop of model and tool calls, its steps and any handoffs. Dome sits at the call boundary and governs those calls without running the work. The agent lifecycle, from provisioned and active to suspended and revoked, is a state kept in the registry, and Dome has no workflow model.

SourceDome Systems, domesystems.ai llms.txt and docs.domesystems.ai llms.txt (concepts/agents/agent-lifecycle)read 2026-09-25

Agents call the gateway, and Clawvisor does not start them. When a held request resolves or a task is approved, denied, expanded or expires, Clawvisor POSTs a callback to a URL the agent supplied. Those callbacks answer the agent's own requests and do not start new work. No event, schedule or channel in Clawvisor starts an agent.

SourceClawvisor, README at github.com/clawvisor/clawvisorread 2026-09-25

The customer's own scheduler, queue or event bus supplies the work, and Dome does not start agents. In Dome's event driven deployment example, an EventBridge schedule fires a Lambda, and retries and concurrency belong to that trigger. Dome governs the tool call the handler then makes. Signed webhooks send Dome's own audit events outward and wake no agent.

SourceDome Systems, docs.domesystems.ai/tutorials/production/deploy-your-agent/event-driven and llms.txt (operate/webhooks)read 2026-09-25

The product gates the agent's calls and indexes none of the customer's documents for an agent to ground its answers on. Chain context extraction pulls identifiers from API responses, but only for Clawvisor's own checks.

SourceClawvisor, clawvisor.com/faqread 2026-09-25

Tool and model calls are governed, but none of the customer's documents are indexed for an agent to ground its answers on. Guards inspect content as it passes through and do not retrieve anything from the customer's corpus.

SourceDome Systems, domesystems.ai llms.txtread 2026-09-25

The state Clawvisor keeps is about the agent's authorization. Chain context verification extracts structural facts (IDs, email addresses, phone numbers) from adapter results and feeds them into Clawvisor's own verification prompts for later steps. Standing tasks keep a grant open until revoked. Clawvisor applies both as rules, and the agent never reads them as context for its decisions. Clawvisor names no memory layer that the agent reads.

SourceClawvisor, README at github.com/clawvisor/clawvisor and clawvisor.com/faqread 2026-09-25

The registry, rules, quotas and audit events are records about agents. The platform and people apply or read them, and the agent never reads them as context to decide. Model Broker pools can cache responses to save cost and latency. Dome names no memory layer.

SourceDome Systems, domesystems.ai llms.txt (platform/registry, platform/broker)read 2026-09-25

A person approves each task's declared purpose before the agent can act. Calls outside that purpose, or that score high for blast radius, are held before they leave the gateway while the agent waits for a decision. High or critical risk tasks need a confirmation step. Sending an iMessage always requires approval, and any service or action can be blocked outright. Pending approvals expire after a set timeout, and people approve from the dashboard or a paired mobile device.

SourceClawvisor, clawvisor.com/faq and /security, and the README at github.com/clawvisor/clawvisorread 2026-09-25

Every tool and model call is checked against Cedar rules at the call boundary before it executes, failing closed with forbid winning by default. Rules can be scoped from the organization and tenant down to a workspace or a single agent, and they ship as versioned code. Guards inspect, redact and constrain content on tool and model connections, and quotas cap spend, tokens or calls. These controls stop an action outright. Decisions are made by policy, and Dome names no step where a person approves an action.

SourceDome Systems, domesystems.ai llms.txt (platform/authorization) and docs.domesystems.ai llms.txt (govern/guards, reference/mcp/quotas)read 2026-09-25

SAML and OIDC single sign on ship for Enterprise orgs. Each agent authenticates to the gateway with its own bearer token, and management API tokens are scoped, revocable and hashed at rest. Every tool call is limited to the task a person approved, with blocked lists of services and actions for least privilege tool access. No SOC 2, ISO or other attestation is published. Clawvisor asks customers to verify its security claims by reading the public source, and its README warns that the software is experimental and has not been audited for security. Clawvisor names no RBAC or ABAC model.

SourceClawvisor, clawvisor.com/faq and /security, and the README at github.com/clawvisor/clawvisor (linked from clawvisor.com)read 2026-09-25

Platform RBAC has built in and custom roles across organization, tenant and workspace scopes, with scoped platform API keys, all kept separate from runtime agent authorization. Every agent authenticates with the credential issued at registration. The end user an agent acts for is verified per agent through OIDC or HMAC instead of being trusted, so both identities reach each Cedar authorization decision and the audit record. Tools and models are reachable only through Gateways an admin grants, with upstream credentials injected at egress, so tool access is held to what is needed.

Dome names no attestation. For security measures the DPA points to trust.domesystems.ai, a Vanta trust center.

SourceDome Systems, docs.domesystems.ai llms.txt (manage/access, reference/permissions, concepts/platform/permissions, connect/agents/delegated) and domesystems.ai llms.txt (platform/access, dpa), plus trust.domesystems.airead 2026-09-25

Every protected tool call is recorded with its action, arguments, policy decision and the task that authorized it. Each record can be replayed and is tied to the person who approved the work, and every purpose declaration and credential injection is recorded too. Activity rolls up by task and approver, so a customer sees which accounts each agent touched, and the web and terminal dashboards show it live. The managed cloud adds extended log retention on the Pro plan.

SourceClawvisor, clawvisor.com/security, /faq and /pricingread 2026-09-25

Every governed action emits an immutable audit event recording the agent, the caller chain and the tool or model, along with arguments, policy version and outcome, in one vocabulary across tool and model calls. Events can be queried from the UI and CLI or through the API and MCP server, exported as OCSF, CEF or CSV, and streamed continuously to a SIEM or exported in scheduled batches.

SourceDome Systems, domesystems.ai llms.txt (platform/audit) and docs.domesystems.ai llms.txt (manage/export, reference/events/catalog)read 2026-09-25

Customers choose between a managed cloud and self hosting inside their own network from the public source, which runs as a gateway service plus a datastore for the vault and audit log. Purpose verification, risk scoring, the vault and the audit trail work the same either way. When self hosted, credentials and traffic never leave the customer's network. Setup guides cover Docker, VPS, container platforms and Cloud Run. Data residency options are planned as future Enterprise work.

SourceClawvisor, clawvisor.com/faq and /pricing, and the README at github.com/clawvisor/clawvisorread 2026-09-25

Four hosted tiers are sold. No hosting region is published, and no tier runs Dome itself in a VPC, in the customer's own environment or self hosted. Tenants are isolated logical environments inside an organization. The governed agents can run on any cloud or runtime, but Dome itself runs as a hosted service.

SourceDome Systems, domesystems.ai/pricing and llms.txt, and docs.domesystems.ai llms.txtread 2026-09-25

No ready made agents, workflows or templates ship with the product. Its fourteen service adapters are integrations, and the installable skill only teaches an agent to route its calls through the gateway.

SourceClawvisor, clawvisor.com/integrations and /faqread 2026-09-25

No ready made agents, workflows or templates ship. Tutorials and field guides walk a customer through building and deploying its own agents. Rules, guards and gateways are objects the customer writes.

SourceDome Systems, domesystems.ai llms.txt and docs.domesystems.ai llms.txt (tutorials)read 2026-09-25

When self hosting, the operator picks the model behind Clawvisor's own AI checks (intent verification, task risk scoring and optional chain context extraction). The choices are Anthropic, OpenAI, Google Vertex AI, Groq or a local Ollama model, set in the setup wizard or config with the customer's own key. On the managed cloud Clawvisor chooses the model. The proxy-lite runtime, which presents Anthropic and OpenAI compatible endpoints to agents, is in preview.

SourceClawvisor, clawvisor.com/faq and the README at github.com/clawvisor/clawvisorread 2026-09-25

The Model Broker puts model choice with the customer. Admins connect providers such as Anthropic, OpenAI and Bedrock, or self hosted models, with their own credentials behind an API compatible with OpenAI and Anthropic. They compose them into pools with routing strategies, failover and caching, and set model choice as routing policy for each call.

SourceDome Systems, domesystems.ai llms.txt (platform/broker, llm-router field guide) and docs.domesystems.ai llms.txt (reference/mcp/models and pools)read 2026-09-25

The platform has its own HTTP API. Agents create tasks and send gateway requests with bearer tokens, and a management API takes scoped, revocable API tokens so Terraform and CI can drive it. A CLI manages the server, services and agents. Clawvisor also serves an MCP server at /mcp with OAuth 2.1, whose tools (fetch_catalog, create_task, get_task, complete_task, expand_task, gateway_request) create and change its core objects. There is no SDK. Agents connect through a skill or plain HTTP.

SourceClawvisor, README at github.com/clawvisor/clawvisor (linked from clawvisor.com) and clawvisor.com/faqread 2026-09-25

Go and Python SDKs, a CLI and a REST API with a published reference drive the platform, with rules simulation at POST /v1/rules/simulate among the endpoints. An MCP server exposes Dome itself, with tools that register agents and attach tools and models, write rules, guards and quotas, and query audit.

SourceDome Systems, docs.domesystems.ai llms.txt (reference/surfaces/api, sdk, mcp and the MCP server reference) and domesystems.ai llms.txt (platform/operator)read 2026-09-25

Evaluation suites test Clawvisor's own intent verification model, with 179 cases across benign tasks, scope creep and adversarial prompts and 249 across all three checks. The results sit in the public repo, and a customer can rerun them. Clawvisor names nothing that tests, scores or optimizes the customer's agent.

SourceClawvisor, clawvisor.com/faqread 2026-09-25

Authorization rules can be simulated against real replayed traffic and synthetic scenarios before they reach production, through the UI, CLI and API, and rules ship versioned through CI/CD. Simulation gates rule changes in the release path and shows how a new rule would decide the agent's calls. Dome names nothing that scores or tests the agent's own output.

SourceDome Systems, docs.domesystems.ai llms.txt (govern/rules/simulate, cli/secure/rules, api simulate) and domesystems.ai llms.txtread 2026-09-25

Clawvisor brokers API calls to services such as Gmail, GitHub and Slack and does not operate a browser or a computer.

SourceClawvisor, clawvisor.com and /integrationsread 2026-09-25

Dome governs agents' tool and model calls and does not operate a browser or a computer.

SourceDome Systems, domesystems.ai llms.txt and docs.domesystems.ai llms.txtread 2026-09-25