Polter Terminal Lets One AI CLI Supervise Others to Keep Overnight Runs Alive

·1 views

A developer has released Polter, a Ghostty terminal fork that adds a supervisory layer for running multiple AI command-line tools like Claude Code, Codex, and Qwen simultaneously. The tool uses an MCP (Model Context Protocol) layer to let one designated AI agent open tabs, type into other agents' terminals, and read their screens. If a worker agent goes idle or stops responding, the supervisor is automatically notified and can intervene; if the supervisor itself goes quiet, Polter nudges it back into action. Worker agents report progress through a shared group chat and task list, addressing common failure points like unexpected exits and inaccurate handoff reports. The project is MIT licensed and available on GitHub, with full end-to-end testing currently limited to Claude Code as the supervisor agent.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Builds Lightweight Go Gateway to Centralize LLM Provider Setup

A developer created Go-Ai, a small open-source HTTP gateway written in Go, to eliminate repetitive LLM provider configuration across multiple projects. The tool acts as a single proxy layer between application backends and AI providers like OpenAI, handling credentials, model name aliases, and streaming without requiring each app to manage those details independently. The project was motivated by the recurring friction of managing provider keys, model slugs, and environment-specific availability every time a new AI feature was tested. Go-Ai is intentionally narrow in scope and is not designed to replace broader frameworks like LiteLLM or LangChain, which the author acknowledges serve different needs. The gateway is best suited for developers running multiple small apps who want a self-hosted, minimal service they can trace from request to response.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Rewriting Tool Descriptions Boosted AI Agent Accuracy from 34% to 100% for $4

An open-source project called Toolmetry found that vague or outdated tool descriptions — not the AI model itself — were causing agents to fail tasks at high rates. By rewriting only the text descriptions that tell agents what each tool does and how to use it, SQLite task success jumped from 34% to 100%, with the entire experiment costing just $4 in API calls. Researchers identified three root causes: overlapping tool descriptions causing wrong tool selection, implied but unnecessary prerequisite steps wasting tokens, and deprecated parameter references leading to confident but incorrect calls. For example, a git server's success rate rose from 75% to 96.7% simply by updating parameter names in its descriptions to match the current API. The findings suggest that tool descriptions function as a contract between an agent and external systems, and ambiguity in that contract reliably produces failures regardless of model quality.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Natural Language Is a Data Interface, Not a Substitute for Semantic Modeling

A two-part technical series from DEV Community explores how an Enterprise Data Discovery Assistant enables business users to query a Customer 360 domain of over 100 million records without knowing SQL or database schemas. The system relies on Snowflake semantic views to make analytical decisions explicit, covering metrics, relationships, time semantics, and access rules, rather than letting a language model guess from raw table names. Before any query runs, the assistant performs a mandatory data-contract check to verify freshness, quality, and trustworthiness of the target data product. Depending on the contract result, the system either proceeds normally or attaches a warning if the data is stale, degraded, or failed. The core finding is that natural language broadens access to well-modeled data but does not replace the need for rigorous data modeling and governance infrastructure.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

CS Student Builds Three Projects That Each Forced Him to Rethink Core Assumptions

A computer engineering student and developer built three distinct software projects this year, each presenting unexpected design challenges. The first replaced a truck-weighing system at a plant, requiring a serial-port data stream architecture with FastAPI and WebSockets, plus a strict PostgreSQL state machine to prevent edits to completed billing records. The second, a campus social app called Epa, tackled privacy-sensitive features like opt-in expiring location sharing and server-controlled ephemeral media rather than client-side deletion. The third, a PWA music player called BeatWave, exposed the complexity of true offline-first design — managing cache conflicts, interrupted downloads, and partial reconnections via IndexedDB and service workers. Across all three projects, the developer found that the hardest problems were not the obvious technical ones, but the hidden assumptions about data flow, user permissions, and connectivity states.

0 comments Read more at DEV Community