Hank (Hankwave)

Whip is a Go-based coding-agent harness distributed as a single binary with no runtime dependency. It runs an LLM tool-use loop for bash commands, reading, writing, editing, and subagents, alongside an interactive Bubble Tea terminal interface. The harness supports parallel tool calls, streaming, background subagents, MCP servers, and provider-routable models with live discovery from provider catalogs; any OpenAI-compatible endpoint can be used as a provider. Prebuilt checksum-verified binaries are available for Linux and macOS on x64 and arm64, and it can also be installed from source with Go.

WolfBench is an open-source Apache-2.0 toolkit and evaluation methodology by Wolfram Ravenwolf for collecting, curating, analyzing, and publishing AI-agent benchmark results. It evaluates agents on Terminal-Bench 2.0 through Harbor orchestration and Daytona sandboxes, preserving configuration metadata, per-task results, token usage, timing, failures, and trajectory traces. Rather than reporting only an average, WolfBench calculates five metrics: the union of tasks solved across runs (Answers), the best single-run score, the mean across valid runs, the worst single-run score, and the tasks solved in every run (Solid). Repeated runs under consistent benchmark, runner, sandbox, agent, model-service, and settings configurations expose variance and distinguish dependable performance from isolated successes. The repository includes a Marimo dashboard that scans local and remote run directories over SSH, deduplicates and classifies runs, filters incomplete or invalid results, downloads data, saves curated JSON snapshots, uploads evaluations and trajectory traces to W&B Weave, and generates interactive HTML leaderboard charts. The core collector and chart generator use only the Python standard library; the dashboard requires Python 3.12+, Marimo, Weave/W&B, and SSH access for initial collection. Benchmark data is licensed CC-BY-4.0.

OpenHands Agent Canvas is a workflow and runtime interface for running coding agents across different execution environments. It can use Claude Code, Codex, or the OpenHands harness and move agent work between local execution, Docker containers, remote virtual machines, and Kubernetes, supporting both interactive development and event-triggered background workflows. The OpenHands platform connects triggers from sources such as GitHub, Slack, Jira, Linear, and schedules to agent workflows that plan, execute, validate, ship, and monitor engineering changes. The page presents Agent Canvas as a preview within OpenHands, with automations for tasks such as pull-request review, issue triage, CI-failure remediation, security work, and documentation updates.

Waku is a native desktop app for working with local coding agents. Built in Rust with GPUI, it keeps projects, sessions, transcripts, and application state on the user's machine, while detecting supported agent CLIs and using each provider's native structured protocol and session continuity. Its interface supports independent project sessions, model and reasoning settings, queued or directed follow-up messages, and Git-backed task rewind with conversation-aware checkpoints. The desktop app communicates with a standalone waku-daemon over an authenticated, versioned WebSocket protocol. The daemon manages task data in SQLite, attachments, provider-native session forks, workspace files, and Git operations; a browser client uses the same generated protocol. Waku is distributed for macOS, Linux, and Windows, with no Waku account or remote service required.

Harness² is an open-source framework for recursively optimizing the system around an AI model while keeping the model weights fixed. At the task level, it uses a contrastive Propose–Probe–Compose process: it proposes edits to an execution harness, probes their effects on the same task, and composes a refined harness from behavioral differences between executions without ground-truth feedback. A second recursion level maintains an improvement harness containing adopted harnesses and evidence-supported lessons for future tasks. Its editable components include system prompts, guardrails, memory, tool guidance, subagents, skills, scripts, and plugins. The repository supports parallel and sequential recursion, OpenCode and Codex executors, and benchmarks including JobBench, LAB, and WorkBuddy-Bench. It runs locally with Linux, Python 3.12, and uv, and is licensed under Apache 2.0; the repository states that it is not an officially supported Google product.