Prompts with names, amounts, IBANs, API keys and project names stream through TALOS, which replaces each sensitive value with a token before it reaches the AI tool. No prompt is stored.
See it work
One prompt with TALOS
Someone asks ChatGPT for help with a client's payment. Here's what TALOS does before the prompt leaves, and when the answer comes back.
A prompt to ChatGPT asks for an email to Maria Rossi confirming a €12,500 payment for contract ACME-204. TALOS finds the name, the amount and the contract number, judged by THEMIS in 41 milliseconds; ChatGPT receives tokens instead; its answer comes back with the real values restored. In the Colchix console, a new rule for project codenames is switched on, live on every device in 3 seconds, and the protection events, without the conversation, are sent to ATHENA as evidence.
TALOS in action
Keep sensitive data private. Keep AI working.
Invisible runtime protection across the AI tools your people and agents already use.
- 01
TALOS · Protect
Your AI gets the task. Not your sensitive data.
Sensitive values become tokens before sending, and the originals are restored in the response.
What you write: Draft an email to Maria Rossi confirming payment of €12,500 for ACME-204. What the AI receives: the same request with tokens in place of the name, the amount and the contract. What you get back: the answer, with the real values restored. No prompt content is stored.
- 02
TALOS · Cover
Protection beyond approved AI
Approved tools, shadow AI and personal accounts, protected the same way in the workflows TALOS covers.
Claude and Notion AI on work accounts, ChatGPT and Gemini on personal accounts, each marked protected.
- 03
TALOS · Decide
You define what stays private
Protection rules tailored to your business, from personal data to your own project codenames.
Protection rules for personal data, financial data, credentials, confidential code and trade secrets switch on, then a company's own rule for project codenames is added.
- 04
TALOS · Every way in
For people, agents and MCP workflows
One runtime protection layer for employees and automated AI workflows alike.
Employees, AI agents and MCP workflows all pass through TALOS on their way to ChatGPT, Claude, Gemini and Notion AI.
- 05
TALOS · Record
See what was protected
Every protection event, without the private conversation, ready for ATHENA's audit evidence.
Protection events with the time, the kind of data and what TALOS did, tokenized or masked, never the prompt.
- 06
TALOS · Keep
Your data. Your region.
Choose where the data Colchix manages is kept, to fit your residency requirements.
The storage region set to Europe; a private VPC and on-premises are on request.
- Stronger compliance foundationsMasking and tokenization keep what AI services receive to a minimum, for your privacy and compliance obligations.
- Data minimization
- Confidentiality
- Privacy by design
Keep your tools. Keep your workflow.No separate workspace, no extra steps: TALOS works where your teams already do.
ChatGPT
Claude
Gemini
Notion AI- TALOS active
The risks of unprotected AI usage
A single prompt can expose more than you think.
Personal data, business secrets and credentials reach AI services through everyday work, by employees and agents alike.
- 01
Approved AI doesn't make every prompt safe
Even an approved tool can receive customer records, financial information or confidential documents that should never be shared.
In ChatGPT, an approved tool, someone asks to summarize Maria Rossi's contract and her €12,500 payment, with a customer records file attached: sensitive information included.
- 02
Personal accounts can carry company secrets
Business information leaves through personal accounts and unapproved tools, outside the environment you manage.
A confidential sales plan is pasted into Claude, on a personal account nobody approved, with a request to improve it for next quarter.
- 03
Your intellectual property can leave with a paste
Confidential code, product plans, credentials and trade secrets can all end up in an AI request.
A snippet from a private repository, with an API key and a database address in it, is sent to an AI tool with the product roadmap, flagged as credentials, confidential code and trade secrets.
- 04
Your privacy obligations don't stop at the prompt
AI interactions can expose information covered by privacy law, customer agreements and security requirements.
A customer's name, email, account details and location go into a prompt, touching privacy requirements, customer commitments and security policies.
Data exposure can become a costly business problem.
Fines & legal costs
Lost intellectual property
Lost trust
TALOS impact
See the difference with TALOS
Less data exposure, AI workflows that keep moving, and protection you can show.
Runtime protection
41 ms
to judge a prompt and mask it, before it leaves
In the bank in our film: a client's name and IBAN, judged by THEMIS and masked by TALOS before ChatGPT saw a thing. Nobody waited.
Prompts stored
0TALOS works on the prompt as it passes and keeps none of it: the protection record says what kind of data, and what was done.
- Masked before sending
- Restored in the answer
- Counts and categories only
- 3 sFor a new rule to reach every device, in the law firm in our film5 minFrom install to protection, live
Who's guarding your data when you use AI?
Let TALOS, your bronze guardian, protect sensitive data at runtime.
While your teams keep moving at AI-speed.
- Any AI tool
- Prompts never stored
- Live in 5 minutes
Colchix governs the AI tools your teams already use, among them ChatGPT, Claude, Gemini, Microsoft Copilot, Cursor, Mistral, Perplexity, DeepSeek, GitHub Copilot, Grok, Notion AI, Hugging Face, Meta AI, Qwen, Ollama.
TALOS FAQs
Your questions about TALOS, answered.
What is TALOS and how does it protect sensitive data?
TALOS is Colchix's AI Data Protection Agent. It detects sensitive information in supported AI interactions and masks or tokenizes it before it reaches the AI service. Where reversible tokenization is used, TALOS restores the original values in the response, so teams keep working without removing and reinserting information by hand.
Does TALOS protect personal accounts and unapproved AI tools?
Yes. TALOS extends protection beyond approved company tools to supported interactions with shadow AI and personal accounts. Protection follows the sensitive information being shared, rather than relying only on whether a tool is approved, which covers the everyday situations where people use AI outside centrally managed accounts.
What types of sensitive information can TALOS protect?
TALOS protects personal data, financial information, credentials and confidential business content, including sensitive code and trade secrets. You can tailor the protection rules to what matters to your business, combining common categories of sensitive data with your own company-specific ones.
Will employees need to use a different AI tool or workspace?
No. TALOS works within the supported AI tools and workflows your teams already use, with no separate chat workspace. Once installed and configured, protection runs in the background. Employees keep working as usual, without cleaning up prompts by hand or switching between platforms.
Does TALOS read or store our prompts?
TALOS analyzes content during the interaction to detect and protect sensitive information; it does not store prompt content. Protection records show the data category, the action and the time, never the private conversation. The Security page explains where processing happens, which metadata is kept and for how long.
Security & deploymentCan TALOS protect AI agents and MCP workflows?
Yes. TALOS protects employee interactions and supported automated workflows, including MCP connections, applying protection before sensitive information reaches the AI service through the configured integration. Coverage depends on the tools and connections in your environment.
How is TALOS different from simply blocking AI tools?
Blocking takes a tool away; TALOS protects the sensitive information inside supported interactions, so teams keep using AI. Masking and tokenization transform what's shared, and reversible tokenization restores the original values in the response. You no longer have to choose between unrestricted sharing and blocking a whole workflow.
How does TALOS fit alongside existing DLP and compliance tools?
TALOS adds a protection layer built for AI interactions, alongside your existing Data Loss Prevention and governance controls. Masking and tokenization minimize the sensitive information shared with AI services, and protection events feed the evidence ATHENA keeps, supporting your compliance work together with your policies, processes and other safeguards.
Explore ATHENAWhat is THEMIS?
THEMIS is Colchix's System One² AI model, the judgement under ARGUS, TALOS and ATHENA. For TALOS, it judges what's sensitive in every prompt before TALOS masks it: a client's name, an IBAN, an API key or a project codename, in milliseconds. It decides on your rules.
Explore THEMISAsk AI about TALOS
Get an independent answer. The question's ready, pick your assistant.
