Fibona AI
Fibona is an open-source team exploring the limits of intelligence through recursive self-improvement (RSI). Its work focuses on AI agents and harnesses that can write, extend, and repair their own code from a minimal core.
At a Glance
- AI researchers studying recursive self-improvement and agent harnesses
- Developers building general-purpose or self-improving agents
- Open-source AI and developer-tool communities
- Individuals who want browser-based, local, or iPhone AI-agent experimentation
AI Tools by Fibona AI
(1)
Fibona Agent
Open Source Self Writing Agent
Discussions
No discussions yet
Be the first to start a discussion about Fibona AI
Latest News
Fibona promoted Fibona Agent's iPhone experience, including a natural-language iPod example.
Fibona introduced Fibona Agent as a minimal, general-purpose self-improving agent with a roughly 200-line core, open-source code, and a free demo.
Fibona announced Tabgent, an open-source browser agent that brings Codex into Chrome.
Fibona published the technical report “Fibona Agent: The Ultimate Harness Bootstrapping Itself From Scratch.”
Products & Services
An open-source, minimal general-purpose self-improving agent that writes its own code as it runs, builds its system step by step, and uses a persistent session, history, and recovery cells to repair failures. It is available as source code, a web/cloud-sandbox demo, and an iPhone experience.
An open-source Chrome extension that puts Codex in the browser sidebar. It can read the current page, compare tabs, explain and annotate PDFs, interact with websites, fill forms, search, and retain conversations associated with URLs.
Market Position
Fibona positions Fibona Agent as more fundamental and more self-directed than conventional ReAct-style agents: instead of a developer-owned loop selecting one tool call at a time, the agent writes the code for its next step and controls how it calls the model and manages context. Its technical report compares the approach with Claude Code, Codex, RLMs, Prime Agent, DeepSeek Harness, and other self-improving-agent research, arguing that Fibona combines persistent code execution with an explicit minimal immutable core that defines the agent's self.
Founding Story
Fibona publicly introduced itself on September 18, 2026 as a project building intelligence that improves itself, compounds over time, and stays open to everyone. The initial vision was to explore recursive self-improvement and minimize human-predefined behavior in AI harnesses; Fibona Agent embodies this by letting an agent write its own next-step code and repair failures from a small immutable core.
Business Model
Revenue Model
Fibona operates a hosted web/cloud-sandbox and iPhone service with account sign-in and daily budgets for model requests; the agent source code itself is available under the MIT License.
Target Markets
- AI researchers studying recursive self-improvement and agent harnesses
- Developers building general-purpose or self-improving agents
- Open-source AI and developer-tool communities
- Individuals who want browser-based, local, or iPhone AI-agent experimentation
- Building customized AI agents and harnesses from a minimal core
- Recursive self-improvement and research into self-modifying AI systems
- Autonomous coding, tool use, package installation, subprocess execution, and workflow construction
- Long-running tasks that need persistent memory, context management, and failure recovery
- Browser research, reading and annotating papers or manuals, comparing tabs, filling forms, and website interaction through Tabgent
- Personal experimentation with AI agents on the web, in a cloud sandbox, on an iPhone, or locally via the repository