Formal vs AI gateways

Model protection, tool permissions, agent context, and direct production access.

AI gateways

CriteriaFormalLiteLLMRunlayerNoma
Agent identity and human attribution

Documented: Agents linked to their operator

Documented: Gateway users, teams, keys and agents

Documented: Connector access and scoped agent accounts

Documented: Agent ownership and runtime identity

Upstream credential handling

Documented: Configured secrets held at the proxy

Documented: Provider credentials and MCP OAuth tokens

Documented: Connector OAuth and managed secret stores

Partial or edition-limited: Native secret injection not documented

MCP server governance

Documented: Remote HTTPS MCP governance

Documented: Remote and supported local MCP aggregation

Documented: Local and remote MCP governance

Documented: MCP discovery and integrated controls

Tool-level permissions

Documented: Per-tool and context policies

Documented: Tool and allowed-parameter permissions

Documented: Tool scanners and AgentGuard decisions

Documented: Identity and context-aware tool policies

LLM traffic inspection

Documented: Model request and response inspection

Documented: Model routing, accounting and guardrails

Documented: Integrated model and session visibility

Documented: Integrated model and session inspection

Prompt-injection detection

Documented: Log-based Insights analysis

Documented: Configured injection guardrails and hooks

Documented: AgentGuard session risk evaluation

Documented: Context-aware injection risk detection

Native database access

Documented: Native database protocol proxy

Not documented: No native database wire-protocol proxy

Partial or edition-limited: Database operations through tools

Partial or edition-limited: Integrated tools, not native SQL proxy

Database masking and result filtering

Documented: Mask fields and filter results

Partial or edition-limited: Guardrails, not native SQL result controls

Partial or edition-limited: Tool content masking, not native SQL

Partial or edition-limited: AI content masking, not SQL result policies

Full comparisonRead Formal vs LiteLLM ›Read Formal vs Runlayer ›Read Formal vs Noma ›

Documented Partial or edition-limited Not documented

Formal publishes these comparisons. Each cell links to the full explanation and vendor sources.

Vendor comparisons

Teams that run Formal

  • Cursor
  • Notion
  • Profound
  • Surge
  • Asana
  • Decagon

Formal enforces policies on agent traffic

Model calls, tools, APIs, and production data share one policy engine.

Get a demo ›Get a quote ›