Mastra

Integrate Intutic governance with Mastra agents.

Mastra is governed on two independent surfaces:

1. LLM egress — point the model client's baseURL at the Intutic proxy (or launch under intutic exec). Every LLM call crosses the proxy and is governed like any other harness.
2. Local tool execution — Mastra tools run as plain objects inside your own Node.js process. No config or hook file can gate them, so the blocking gate ships SDK-side, in @intutic/gate/mastra.

How it works

The mastra adapter is detected when package.json declares an @mastra/core dependency (any version). Like every other SDK-gated framework in this codebase, it writes a .env.intutic file with proxy base-URL env vars plus a comment block pointing at the SDK gate.

Setup

1. Initialize Intutic

intutic init

✓ Detected harnesses:
  • mastra → .env.intutic

2. Route LLM traffic through the proxy

Source the generated env file, launch under intutic exec, or set the model provider's baseURL explicitly — Mastra reads whatever base URL the provider you construct is given.

3. Gate local tool execution (SDK)

npm install @intutic/gate

Mastra exposes a documented pre-execution veto point — an Agent's hooks.beforeToolCall — which intuticHooks() implements:

import { Agent } from '@mastra/core/agent'
import { Gate, install } from '@intutic/gate'
import { intuticHooks } from '@intutic/gate/mastra'

install(new Gate({ workspaceId: process.env.INTUTIC_WORKSPACE_ID }))

const agent = new Agent({
  id: 'my-agent',
  name: 'My Agent',
  model: 'openai/gpt-5-mini',
  tools: { deleteTool, shellTool },
  hooks: intuticHooks(),
})

On deny, beforeToolCall returns { proceed: false, output } — Mastra skips the tool's real execute entirely and hands output back to the model in its place. The default output shape is { error: true, message, code, incidentId? }; if one of your tools declares a strict Zod outputSchema, pass a denialOutput shaper so the refusal still validates:

intuticHooks({
  denialOutput: ({ toolName, refusal }) => ({ deleted: false, reason: refusal.message }),
})

This hook runs for every tool in the agent's assembled tool dictionary — natively defined tools and MCP-sourced tools (registered through @mastra/mcp) alike, since both are merged into the same tool record before hooks wrap it.

4. Trace attribution

import { intuticHeaders } from '@intutic/gate'

// pass intuticHeaders({ sessionId: runId, harness: 'mastra' }) as defaultHeaders
// on whatever client/provider you construct for the model

What gets written

Same shape as every other SDK-gated framework's .env.intutic — proxy URLs plus a pointer at @intutic/gate/mastra's intuticHooks().

Known bypass: per-call hooks override, not merge

This is Mastra's own documented behaviour, not an Intutic defect — but it is a real gap operators need to know about. new Agent({ hooks: intuticHooks() }) installs this gate at the agent level. However, if any caller of that agent's .generate()/.stream() passes its own hooks option — even {}, or a hooks object with no beforeToolCall — Mastra replaces the agent-level hooks wholesale for that call. There is no merging: the agent-level intuticHooks() silently does not run, with no error and no warning.

If your application code passes call-level hooks anywhere, compose this gate into that call's hooks yourself:

const intutic = intuticHooks()
await agent.generate(prompt, {
  hooks: {
    beforeToolCall: async (ctx) => {
      const denied = await intutic.beforeToolCall(ctx)
      if (denied) return denied
      return myOwnBeforeToolCall(ctx)
    },
  },
})

Confirmed against a real install (@mastra/core@1.59.0): Agent.getConfiguredToolHooks()'s own doc comment states "Run-level hooks override these ... callers that need to preserve the configured hooks must read and compose them explicitly." See docs/TECH_DEBT.md TD-380 and @intutic/gate/mastra's module doc for the full record.

What the adapter does NOT do

Same structural gaps as every SDK-gated framework — see LangGraph's "What the adapter does NOT do". In short: there is no sync-daemon hook file, argument-level gating requires wiring the SDK into your own agent code, and x-intutic-harness attribution is client-supplied, not authorization. On top of that structural family, Mastra adds the per-call-hooks-override bypass above, which is specific to this framework's own hook design.

Config details

Property	Value
Harness type	mastra
Config file	.env.intutic
Detection	@mastra/core in package.json (dependencies, devDependencies, or peerDependencies)
Format	Shell environment variables
Write strategy	Atomic (write to .intutic-tmp, then rename)
Tool gate	SDK-side (@intutic/gate/mastra's intuticHooks(), Mastra's Agent.hooks.beforeToolCall) — no sync-daemon hook file; see the per-call-hooks-override bypass above