Mert Köseoğlu is a software engineer and the founder of Context Mode. His projects work on the infrastructure around AI agents: controlling context size, running issue-based coding workflows and connecting assistants to browser sessions. His earlier engineering roles include Qooper, Planhat and consulting work.[2][3][4][5]
Start with Context Mode →
X Career
Köseoğlu describes a background in full-stack engineering and technical leadership, with work on customer platforms, mentoring software and enterprise systems. His portfolio dates the founding of Context Mode to March 2026. His consulting work at G2i included AI testing infrastructure for a client project at OpenAI.[2]
Contributions
Keeping tool output out of the conversation
Context Mode addresses a source of context growth that can be easy to miss: large tool results are repeatedly included in later model requests. Its design executes and indexes material outside the conversation, then brings back selected information through search and routing. The repository also describes session continuity across supported agent platforms. The mechanism is selective retrieval and controlled tool-output handling; a benchmark reduction on one workload should not be read as a universal saving.[3]
Issue orchestration and browser access
Hatice polls an issue tracker, creates isolated workspaces and dispatches coding agents to work on each issue. Its workflow file describes the tracker, workspace setup and task instructions, while retries and reconciliation manage failures over time. Browsir AI solves a separate integration problem by connecting an agent to a local Chrome instance through the Chrome DevTools Protocol. These tools operate at the harness level: they determine the environment and tools an agent receives, and therefore what actions it can take.[4][5]
Tools, writing & talks
Tools
Start here
Context Mode
github.comA context-management layer that handles tool output outside the conversation and retrieves selected information for coding agents.
Read the architecture and context examples first to see whether large tool output is your bottleneck. Installation depends on your coding-agent platform; inspect its routing and execution behavior before enabling it.
Creator and maintainer[3]Original source- AI Phone Agentgithub.com
A proof of concept combining FreeSWITCH, transcription, a language model, synthesized speech and human call transfer.
Repository owner[6]Original source - SeclawAIseclawai.com
Builds secure autonomous AI agents on top of OpenClaw with a 60-second setup.
Catalog workOriginal source - Browsir AIgithub.com
An MCP server and CLI that connects agents to a local Chrome browser through CDP.
Repository owner[5]Original source - Haticegithub.com
An issue-driven orchestrator with isolated workspaces, coding-agent dispatch, retries and reconciliation.
Repository owner[4]Original source