The agent control plane
Security infrastructure for AI agents.
Let your agents act on real systems, with your policy, your approvals and your audit trail.
Titan · private preview
One path for every action an agent takes.
If Titan can’t decide, nothing runs.
- 01
The agent asks
- 02
Policy decides
- 03
A person approves
- 04
The gateway adds the key
- 05
The ledger records
What teams say
Controls were built for people, not for agents.
“I don’t know how many agents we have or what they can reach.”
“Our agents run on someone’s personal token.”
Control, no standing secrets, and proof.
Decide every action before it runs.
Allow, approve or deny, per tool and resource.
The agent never holds a key.
Credentials stay in the vault.
Every step, sealed to the one before.
Hash-chained, signed, exportable to your SIEM.
One policy for agents on laptops and in the cloud.
laptop · codexrules v12 · signed
$codex cat ~/.ssh/id_ed25519✕refused · rule: no reads under ~/.ssh$codex npm test✓allowed · 14 passed · 2.1s$codex git push --force origin main✕refused · "Force pushes go through a reviewer." same rules on every laptop · signed by acme
titan cloud · support-agentacme · production
$support-agent zendesk.reply ticket 48122✓allowed · grant 30s$support-agent stripe.refund 48122 $129.00 needs approval · payments lead✓approved by raj · ledger #7c90…e3$support-agent crm.export customers✕denied · outside the agent's scope
Runs withClaude CodeCodexKiroCursorOpenAI Agents SDKAny MCP clientYour own harness
Eleven capabilities live. Two in early access.
Titan in detail- Agent identity and registry
- Per-action policy
- Human approval, bound to the action
- Credential vault and 30-second grants
- Reviewed tool catalog
- Agent Ledger, hash-chained
- Audit export to your SIEM
- Kill switch
- Environment freeze
- Coding-agent governance
- Titan Cloud and Dedicated, custom domains
Early access
- Durable sessions across machines
- Defender
Our cloud, or a dedicated Titan. Same policy.
- Titan Cloud
Hexel Studio's cloud, isolated by organization
- Titan Dedicated
A Titan for one organization, in our cloud
- 01
9
levels you can stop an agent at
- 02
30s
longest life of an execution grant
- 03
4
Defender rules, each revokes the session
- 04
0
credentials handed to an agent
Works with your existing stack.
- Slack
- Google Sheets
- MCP servers
- HTTP APIs
- GraphQL APIs
- Any MCP client
mcp.json
{
"mcpServers": {
"titan": {
"url": "https://api.hexelstudio.com/v1/titan/agent/mcp",
"headers": { "Authorization": "Bearer $TITAN_RUNTIME_TOKEN" }
}
}
}Put your agents to work, safely.
Tell us what your agents do. A person replies within one business day.