Releases·Tool Release
tiny-agent: modular Node.js coding-agent harness exposes the core agent loop with 11 local tools
A GitHub project 'tiny-agent' offers a small, modular coding-agent harness for Node.js that exposes the core agentic loop without framework abstraction. It targets developers wanting a minimal, inspectable foundation for coding-agent tools.
41 engagement1 sourceFri, Aug 28, 2026, 01:50 PM
AI Overview
- tiny-agent (GitHub, MIT) is a small modular coding-agent harness for Node.js 20+ that shows the core agent loop without hiding it behind a framework.
- It ships 11 local tools — read, write, edit, bash, grep, glob, delete, patch, check, fetch, todo — with tool specs in tools/{name}/{name}..
- It runs against any OpenAI-compatible Chat Completions provider via BASE_URL, with configurable MODEL and FALLBACK_MODELS.
- Safety includes symlink-aware workspace containment, step limits, tool-output truncation, and approval prompts for writes and bash outside a git repo; authors state it is not a sandbox.
Entities
Related
Products·launched
NiftyAgent – single-file coding agent CLI under 500 lines
NiftyAgent is a minimal coding agent CLI implemented in under 500 lines of Node.js with no framework or library dependencies. It runs an LLM agent loop with a small set of file read/edit tools, letting developers use it as a baseline for experimenting with context management and task completion strategies.
NiftyAgent is a minimal coding agent CLI implemented in under 500 lines of Node.js with no framework or library dependencies. It runs an LLM agent loop with a small set of file read/edit tools, letting developers use it as a baseline for experimenting with context management and task completion strategies.
1 engagement1 sourcehackernewsTue, Sep 22, 2026, 03:36 PM
Products·launched
Zeroparams: coding agent with zero parameters, human in the loop
Zeroparams is an art project that replaces an LLM in an OpenAI-compatible coding agent with a human. A custom VSCode extension intercepts completion requests and sends them to a chat window, where the human can edit files and run CLI commands via tool calls. It explores the interface between programmers and models, with real subscriptions.
Zeroparams is an art project that replaces an LLM in an OpenAI-compatible coding agent with a human. A custom VSCode extension intercepts completion requests and sends them to a chat window, where the human can edit files and run CLI commands via tool calls. It explores the interface between programmers and models, with real subscriptions.
1 engagement1 sourcehackernewsTue, Aug 18, 2026, 11:54 AM
Products·launched
Faber: open-source coding agent using code graph to cut context and cost
Faber is an open-source AI coding agent for the terminal that uses a code graph to find relevant files and symbols before querying LLMs, reducing context, cost, and search time. It integrates with Claude, OpenAI Codex, and local models, and uses prompt caching and token-conscious output to lower API costs.
Faber is an open-source AI coding agent for the terminal that uses a code graph to find relevant files and symbols before querying LLMs, reducing context, cost, and search time. It integrates with Claude, OpenAI Codex, and local models, and uses prompt caching and token-conscious output to lower API costs.
1 engagement1 sourceredditSun, Aug 23, 2026, 08:59 PM
Products·launched
Polyglot: a terminal coding agent that tolerates malformed tool calls from small local models
Polyglot is a terminal coding agent designed to run against a local Ollama server. Instead of relying on a provider's native function-calling API, it describes tools in the system prompt and parses tool calls as text, recovering from common format errors. This lets people running 3B-8B local models use an agent loop that doesn't stall when the model improvises the tool-call envelope.
Polyglot is a terminal coding agent designed to run against a local Ollama server. Instead of relying on a provider's native function-calling API, it describes tools in the system prompt and parses tool calls as text, recovering from common format errors. This lets people running 3B-8B local models use an agent loop that doesn't stall when the model improvises the tool-call envelope.
1 engagement1 sourceredditTue, Sep 1, 2026, 08:06 PM
Products·building
CleanSlate: open-source coding agent with editor, CLI, and SDK
CleanSlate is an open-source coding agent that understands codebases, edits files, runs commands, browses the web, and verifies its own work. It offers a desktop editor, CLI, and SDK, and supports multiple model providers via user API keys. It solves the problem of vendor lock-in and provides flexibility for developers building custom tools.
CleanSlate is an open-source coding agent that understands codebases, edits files, runs commands, browses the web, and verifies its own work. It offers a desktop editor, CLI, and SDK, and supports multiple model providers via user API keys. It solves the problem of vendor lock-in and provides flexibility for developers building custom tools.
1 engagement1 sourceredditFri, Jul 31, 2026, 04:07 PM