New · a window inside your Talaix account

Agent Access — governed AI access to your evidence

Give AI agents — Claude, Cursor, ChatGPT or any MCP client — controlled, recorded access to Talaix evidence. You decide which agent may call which tool, approve the sensitive calls, and read every action in a tamper-proof audit trail.

Try the tools now Sign in to govern access
What it is

A governance layer above the connection

A tunnel only moves packets. Agent Access decides who may reach which tool, what needs a human, and what is written down.

Connector

A small, outbound-only agent on your network. No open ports, no firewall changes. It dials out to the Talaix gateway and fails closed if the link drops.

Gateway

The policy point. It authenticates the agent, enforces per-tool permissions and rate limits, and holds calls that need a human.

Audit trail

Every call — who, which tool, what data, what result — is recorded in a tamper-proof SHA-256 hash chain you can verify and export.

Why

Agents want your data. You want proof of what they did.

The problem

AI agents are starting to act, not just chat — querying data, drafting disclosures, exporting packs. Opening your systems to them means exposing them, or moving your data into a cloud you don't control.

Agent Access

Keep your data where it is, give each agent a named identity, scope it to exactly the tools it needs, approve the sensitive ones, and keep a proof trail of every call.

How it works

Three steps to governed access

1. Connect

Install the connector on your network, or point an agent straight at Talaix's MCP endpoint.

2. Set rules

Choose which of the 9 Talaix tools each agent may call, and which calls need a person's approval.

3. Approve & audit

Watch the approval queue, read the audit log, and switch any connector or agent off instantly.

Inside your account

A window inside the platform you already use

Agent Access sits next to Dashboard, Reports, Maps and Intelligence — the same Talaix design, the same sign-in, the same billing.

Illustrative window. The first governed tool is Talaix's own evidence engine — the same MCP tools you can try right now below.

Features

Everything in one control surface

Agent identity

Every agent has a name inside your organisation — access is tied to it, not to a shared key.

Per-tool permissions

Allow analyse and hazards, deny evidence export — tool by tool, agent by agent.

Human approval

Mark a tool as require approval; sensitive calls wait in the queue for a person.

Tamper-proof audit

A SHA-256 hash chain per tenant, re-verifiable on demand and exportable as JSONL/CSV.

Data redaction

Secrets and sensitive fields are stripped before results reach an agent, and before storage.

Rate limits

Caps are enforced before a call leaves the gateway, so no agent can exhaust the engine.

Kill switch

Disable a connector or an agent within one policy push.

Outbound-only

The connector opens no inbound listener — nothing new is exposed to the internet.

Security by design

Nine properties, enforced by the engine and verified end-to-end:

  • Deny-by-default — nothing is reachable until a policy allows it.
  • Outbound-only connector — no inbound listener, no firewall changes.
  • One audited event per call, chained with SHA-256 and re-checkable on demand.
  • Rate limits enforced before a call leaves the gateway.
  • Redaction runs before results reach an agent and before they are stored.
  • Human approval gates any tool a policy marks as sensitive.
  • Kill switch disables a connector or agent in one policy push.
  • Fail-closed — if the tunnel drops, execution stops.
  • Tenant isolation — no read path returns another organisation's data.
Try it now

The evidence engine is live today

Talaix's engine is already an MCP server. Point any MCP client at it and call the tools immediately — no sign-up needed for read-only analysis. Governed access (permissions, approvals, audit) is the layer on top, managed from your account.

Add the server to your MCP client

Paste this into your client's mcpServers configuration:

{
  "mcpServers": {
    "talaix": {
      "url": "https://talaix.com/api/mcp"
    }
  }
}
talaix_analyze talaix_hazards talaix_verify_asset talaix_insurance_profile talaix_mapcheck talaix_sustainability_frameworks talaix_sources talaix_briefs talaix_brief

Full setup guide — add Talaix to Claude, Cursor, VS Code and more →

What this is

  • Identity-bound access for AI agents.
  • Per-tool authorization and rate limits.
  • Human approval for sensitive calls.
  • Data redaction before results reach an agent.
  • A tamper-proof audit trail and a kill switch.

What this is not

  • Not a generic tunnel — the connection is the easy part.
  • Not an AI model and not a chatbot.
  • Not a data store for your evidence.
  • Not a replacement for your own identity provider.

Ready to govern your agents?

Try the tools now with no account. When you are ready for named agents, per-tool permissions and an audit trail, sign in — Agent Access is part of your Talaix account.

Try the tools now Sign in to govern access