Find the open-source solution that fits.

pi-mem stores durable facts, daily logs, notes, and a scratchpad as plain Markdown, with tools for reading, writing, and keyword search. It can inject long-term memory, open tasks, and the latest daily logs into each agent turn while leaving older material available on demand. Installation targets the pi coding agent.

Authenticate once with OAuth2 or an API key, then run agent commands through a proxy that injects fresh credentials without exposing them to the child process. Authsome centralizes encrypted storage, token refresh, provider accounts, and headless access for CI or background workers. It can run as a local CLI or a self-hosted Docker daemon; the CLI requires Python 3.13 or newer.

Deploy the same agent through Telegram, Discord, Microsoft Teams, a terminal, or desktop and mobile companion clients. Talon supports interchangeable Claude, Codex, Kilo, OpenCode, and OpenAI Agents backends, along with MCP tools, hot-reloadable plugins, persistent goals, and scheduled background agents. Source installation requires Node.js 24 or later, while releases also provide standalone binaries for several platforms.

picocode is a compact Rust coding agent for interactive sessions, shell pipelines, automated recipes, and focused work such as CI fixes or small codemods. It supports multiple hosted model providers as well as Ollama, and includes filesystem, search, and shell tools. Destructive file operations and command execution require confirmation by default, and the agent can also be embedded as a Rust library.

LoopTroop turns a complex coding ticket into an interview-led plan, small execution beads, isolated worktree changes, and a final review. Its LLM council compares planning drafts, while failed implementation attempts can reset and retry with fresh context. It is designed for long-running feature work rather than quick edits, and the project recommends running unattended agents inside a disposable VM or sandbox.

Untether bridges coding agents running on your own computer or server to a Telegram bot. From a phone, you can send text or voice tasks, stream progress, switch projects and sessions, transfer files, and respond when an agent needs input. It supports Claude Code, Codex, OpenCode, Pi, Gemini CLI, and Amp, but approval controls, plan mode, and cost reporting vary by engine.

OpenCastle generates project-specific agents, skills, rules, and MCP settings for tools such as Copilot, Cursor, Claude Code, OpenCode, Windsurf, and Codex. You can use the team interactively in an IDE or define a Convoy for parallel batch work, with isolated Git worktrees, resumable execution, validation gates, and an observability dashboard.

AgentTier runs isolated, persistent sandboxes for AI agents and coding assistants on Kubernetes. Each sandbox keeps its workspace on a PVC, applies network and resource policies, and exposes an authenticated terminal, file, command, and agent-invoke surface. The platform includes lifecycle management, templates, governance, audit events, OIDC, and optional gVisor isolation, and requires Kubernetes 1.27+, Helm 3, and compatible NetworkPolicy and CSI storage.

Nex turns agent conversations and connected business sources into a shared knowledge graph, so facts recorded in one tool can be recalled from another without repeated explanations. Its setup command detects supported coding platforms and installs hooks, MCP integration, and slash commands, while a hosted MCP endpoint supports other clients. Using the service requires a Nex account and any desired data integrations.

Describe a software product, review the generated specification, and let a team of specialist agents handle architecture, implementation, testing, review, and deployment. Reversible work can flow through one plan gate, while migrations, payments, authentication, and other high-impact changes add deliberate approvals. GreatCTO targets solo founders and one-person engineering organizations rather than multi-developer teams.

SUPER evaluates whether LLM agents can configure environments, execute code, and answer questions drawn from real machine-learning and NLP research repositories. Its dataset is divided into Expert, Masked, and AutoGen task sets, with code for running agents and scoring results. Local mode executes repository code directly on the machine, while Docker and Modal backends offer safer isolation and concurrent benchmark runs.

claudebox launches a Claude Code project through bubblewrap on Linux or sandbox-exec on macOS. It shadows most of the home directory, mounts the project parent read-only, and blocks sensitive runtime sockets by default, with opt-in SSH, GPG, and XDG access. The project explicitly says this is not a complete security boundary; Linux is stable and macOS experimental.

QQcode is a conversational command-line coding assistant that reads project context and can search, edit, and run commands through a stateful terminal. It includes slash-command and path completion, task tracking, configurable model and permission settings, and support for macOS and Linux, with Windows marked experimental.

Brood Box launches Claude Code, Codex, and other agents inside ephemeral microVMs backed by KVM or macOS Hypervisor.framework. A copy-on-write workspace snapshot keeps real files untouched until you inspect and accept the resulting diffs, while egress profiles and selective secret forwarding reduce exposure. The project is explicitly experimental.

codex-profiles wraps Codex's CODEX_HOME support so personal, work, school, or client profiles keep separate authentication, settings, sessions, connectors, plugins, caches, and logs. It launches both the CLI and macOS Desktop app without reading or copying auth.json; its optional parallel Desktop instances remain experimental and do not provide full operating-system isolation.

Wit lets agents declare planned work, reserve individual functions or types, and see overlapping intents, lock intersections, and dependency conflicts before editing. A local daemon shares the state across tools, while accepted interface contracts can be enforced at commit time. Version 1 is single-machine, requires Bun, and supports TypeScript, JavaScript, and Python; most conflicts remain advisory.

AgentLint audits the context a coding agent reads, including AGENTS.md, CLAUDE.md, Cursor rules, and Copilot instructions, alongside repository conditions such as documented build commands, CI, tests, and oversized files. It runs as a standalone CLI in any Git repository, with an optional Claude Code command for teams that want the checks inside their agent workflow.

Start with a rough software idea, refine it through a design conversation, and move into implementation, or point the tool at an existing project for repeated improvement cycles. It tracks experiments locally, evaluates each change before keeping or reverting it, and can run through Claude Code, Bob Shell, or OpenAI Codex.