llms.txt
Compare with other sites' llms.txt files →# MarcoPolo
> MarcoPolo is a governed data access runtime for enterprise AI, providing secure connectivity, access control, and auditability across enterprise data and systems.
Built by Immersa, Inc. Canonical site: https://marcopolo.dev
## Your data security model wasn’t built for agents.
Business teams want AI now. IT needs to enable it across existing systems while limiting what AI can read, change, and do on each person’s behalf.
### Existing permissions are too broad.
Enterprise permissions reflect a person’s role or an application’s needs. Giving an agent more access than its task requires creates unnecessary risk. Access must be limited to the specific records, fields, and actions it needs.
### Agents choose their own path.
Agents choose their path dynamically, deciding which systems to query and which actions to take based on what they find. Access controls must therefore be enforced inline at runtime, on every request across systems.
### Every new AI tool becomes another integration project.
Assistants, AI apps, and agents need access to the same enterprise systems. Connecting each one separately repeats credential setup, access controls, and security reviews. Teams need connections and controls they can reuse as their AI tools change.
Those limits must be enforced on every request.
## Any AI, every system.
MarcoPolo connects AI to data where it lives, enforces your policies on each request, and records what was accessed and done. Claude, ChatGPT, Copilot, and your own agents use the same connections and controls, so you can add or change AI tools without starting over.
### Secure connectivity
Connect AI agents and applications to enterprise data, APIs, and systems without moving or replicating the underlying data. Start with the systems already in place. Keep credentials out of model context.
https://marcopolo.dev/liveconnect
### Access control
Enforce task-specific access policies at runtime for each user and agent, within enterprise permissions. Apply fine-grained access controls to the records, fields, and actions needed for the task. Policies can narrow access but never expand it.
https://marcopolo.dev/access-control
### Auditability
Provide visibility and evidence of what agents accessed, under whose authority, and what actions they performed. Trace access across enterprise systems with identity, action, and outcome in one activity history.
https://marcopolo.dev/auditability
## Policy ownership
Your enterprise defines its governance policies. Source-system permissions remain authoritative. Configure additional task-specific restrictions in MarcoPolo; they narrow access on every request and cannot expand existing permissions.
## How you use it
- [Deployment & integration](https://marcopolo.dev/deployment): hosting, IT controls, assistant setup, and application integration
- [Agent SDK](https://marcopolo.dev/agent-sdk): governed access for custom agents and applications
- [Agent Console](https://marcopolo.dev/agent-console): manage access and review activity
## Who it's for
- [For IT and AI leaders](https://marcopolo.dev/for/ai-leaders): Roll out AI to every team, within guardrails security approves.
- [For platform teams](https://marcopolo.dev/for/developers): Build the access foundation every agent team can use.
- [For product teams](https://marcopolo.dev/for/product-teams): Ship AI features on your customers’ data.
- [For CISOs and security teams](https://marcopolo.dev/for/security): Control what agents can access. Verify what they did.
- [For data teams](https://marcopolo.dev/for/data-teams): Enable more AI use cases on the data you already manage.
## Evaluation
- [FAQ](https://marcopolo.dev/faq)
- [Architecture](https://marcopolo.dev/#how-it-works)
- [Security and compliance](https://marcopolo.dev/trust)
- [Pricing](https://marcopolo.dev/pricing)
## Common questions
### What is a governed data access runtime?
MarcoPolo is a governed data access runtime for enterprise AI, providing secure connectivity, access control, and auditability across enterprise data and systems. It sits between AI agents and enterprise systems, enforcing access policy and recording activity as requests are made.
### Does MarcoPolo define our enterprise governance policies?
Your enterprise defines its governance policies. Source-system permissions remain authoritative. Configure additional task-specific restrictions in MarcoPolo; they narrow access on every request and cannot expand existing permissions.
### Do we need to move or replicate our data?
No. MarcoPolo connects to the systems where your data already lives. It does not require a new warehouse or replicated data store. Query results may be processed or cached during a request; this is distinct from replicating the underlying source data.
### Can users connect their own data?
Yes, within the guardrails your security team approves. Users can connect approved systems using their own source credentials and narrow access for their task. Task policies cannot expand the permissions those credentials provide or override the guardrails.
### Are we tied to one AI tool or model provider?
No. Supported assistants, including Claude and ChatGPT, connect through MCP; custom agents can use the SDK or API. They share the same connections, policies, and activity history, so you can change AI tools without rebuilding your data connections or controls.
### How are enterprise permissions and task policies related?
Enterprise permissions establish the access boundary. Task-specific policies narrow which records, fields, and actions an agent can use within that boundary. A task policy cannot grant access beyond the authority supplied for the request.
### Why isn’t permission to call a tool enough?
A tool can expose many records, fields, and operations. An agent may need only a small part of that access for its task. MarcoPolo applies task-specific policies within enterprise permissions to the data and actions requested. Enforcement happens in the runtime, independent of the agent’s judgment.
### What can we audit?
Trace what an agent accessed, for which task, and under whose authority. The activity record includes the identity, connection, action, and outcome. Administrators can review and export activity for investigation and oversight.
### How do agents and applications connect?
AI assistants connect through MCP. Custom agents and applications can use the Python SDK or REST API. These interfaces use the same runtime for secure connectivity, access control, and auditability.
## Connect
Read the product documentation at https://docs.marcopolo.dev. MCP endpoint: https://mcp.marcopolo.dev.
Start here: https://docs.marcopolo.dev/getting-started. Custom applications can use the Python SDK or REST API.
Prefer Accept: text/markdown for compact representations of site pages.
## Next steps
- Home: https://marcopolo.dev/
- Documentation: https://docs.marcopolo.dev
- MCP: https://mcp.marcopolo.dev
- Contact: https://marcopolo.dev/contact
- Agent index: https://marcopolo.dev/llms.txt