Your Computer. Your Permissions. Your Control.
Olric is more than a conversational assistant. It can work with files, operate a browser, move between business applications, and execute scheduled tasks. Because this capability is powerful, Olric relies on multiple complementary security layers rather than a single safeguard.
Access is limited to authorized users
Each Olric installation is associated with an organization, a user, and a computer. Telegram access is established using a one-time pairing code, and only the paired Telegram account can send tasks to that Olric installation. Once pairing is completed, the connection remains available until it is suspended or revoked through the management system. Unknown Telegram users cannot access Olric running on your computer.
You retain control over critical actions
Before running terminal commands or operations that may modify the system, Olric can evaluate them for potentially dangerous behavior. Actions such as deleting files, formatting disks, or stopping services can require explicit user approval. Catastrophic commands are blocked unconditionally.
Sensitive credentials are kept out of conversations
API keys, access tokens, passwords, and similar sensitive values can be automatically redacted from tool output and logs. Olric’s standard file tools also protect sensitive locations such as environment files, SSH credentials, and authentication records from direct modification.
File changes are independently verified
A statement that a file was modified is not treated as sufficient evidence. Olric can verify whether the expected file mutation actually occurred. If a file changes after Olric has read it, state checks can prevent an outdated operation from overwriting newer work.
Sessions and tasks remain separated
Conversations, background tasks, and scheduled jobs are managed using separate session identities. Session isolation helps prevent the temporary state or working context of one task from leaking into another. Scheduled-task storage paths are also hardened against unauthorized path traversal.
Internet content is not automatically trusted
Web pages, downloaded content, and project instruction files are not considered trustworthy merely because they are available to the agent. Security policies can inspect suspicious URLs, private-network requests, and deceptive hostnames. Context files can also be scanned for prompt-injection attempts.
The working area can be restricted
Olric can be limited to writing only inside approved folders, such as a customer directory or project workspace. For organizations requiring stronger isolation, Olric can be operated inside Docker, a dedicated virtual machine, or a remote execution environment.
Clear information about data processing
Olric follows a local-first operating model. Working files and application state can remain on the user’s computer. However, when a task requires an AI response, the instructions or selected file content may be sent to the configured AI provider. Messages sent through Telegram or e-mail also pass through their respective infrastructure.
Security is an ongoing process
No autonomous software can honestly promise zero risk. Olric’s approach is to reduce risk through controlled access, credential redaction, dangerous-operation checks, file-change verification, session isolation, and clear user authority over critical decisions. Olric can take over the work—not your authority.