talos-kernel
Talos is an open-source, self-hosted AI agent for terminal and Telegram that places a deterministic permission kernel between a language model and real-world tool effects. Its core principle is that the model proposes actions but never decides whether they are authorized.
At a Glance
- Individual developers and technical operators
- Security-conscious self-hosters
- Teams that need auditable agent actions on developer or operations machines
- Organizations handling sensitive local files or credentials
- +1 more
AI Tools by talos-kernel
(1)
Talos
Self Hosted AI Agent Framework
Discussions
No discussions yet
Be the first to start a discussion about talos-kernel
Latest News
Talos 0.19.19-alpha adds append-only Telegram completion receipts and improves callback acknowledgement and error diagnostics.
Talos 0.19.18-alpha fixes UTF-8 decoding for server-sent event model responses, including Ollama-shaped streams.
Talos 0.19.17-alpha improves cancellation and fallback-provider state, bounds work announcements, and adds regression coverage.
Talos 0.19.16-alpha makes empty API streams and stream errors remain typed failures through the fallback chain and improves channel backoff.
Products & Services
MIT-licensed, self-hosted Python AI agent and deterministic permission kernel. It accepts instructions through terminal and messaging channels, consults configured language-model providers, and executes only effects authorized by the kernel.
Market Position
Talos positions itself as a narrower, security-first alternative to broad agent platforms such as OpenClaw and Hermes. Its differentiators are a deterministic deny-by-default permission kernel, exact-argument capability tokens, sandbox refusal when isolation is unavailable, and tamper-evident audit records; the project explicitly concedes that OpenClaw and Hermes offer broader channels, browser automation, skills, and everyday automation.
Leadership
Founders
kurdman_007
Publicly identified on Hacker News as Talos's maker/creator; described building Talos to hand an agent a terminal without handing it the operator's trust model. No real name or prior-company biography was published in the sources reviewed.
Executive Team
kurdman_007
Creator / maintainer
The public maker of Talos on Hacker News and the author of the project's public comparison page; no legal name or conventional executive biography was published.
Founding Story
The creator says Talos was started because existing agent approaches either interrupt users with constant confirmations or run behind model instructions and blocklists. The initial vision was a small, deterministic kernel that authorizes every read, write, execution, and delegation effect individually, using exact-argument capability tokens rather than relying on the model's judgment.
Business Model
Revenue Model
The project is distributed as MIT-licensed open-source software for self-hosting. The reviewed sources describe installation, source code, and model-provider configuration but do not document a paid subscription, hosted service, API billing model, or other revenue stream.
Target Markets
- Individual developers and technical operators
- Security-conscious self-hosters
- Teams that need auditable agent actions on developer or operations machines
- Organizations handling sensitive local files or credentials
- Users who want an agent with shell access without granting the language model direct authority
- Giving an AI agent controlled access to a developer terminal
- Coding, debugging, testing, and bounded browser-assisted development work
- Running a self-hosted assistant through Telegram or a terminal session
- Security-sensitive automation where every effect must be explainable and auditable
- Delegating isolated coding or review jobs to confined workers
- Reading local images, audio, video frames, PDFs, and office documents without sending them to a third-party parsing service