Show HN: Podiom – durable sessions, scheduling and goals for local Claude/Codex Podiom, an open-source, local-first workspace for Claude Code and OpenAI Codex, launched on Hacker News, enabling users to give AI agents durable context, schedule work, and pursue goals over time while keeping sessions, projects, tasks, and decisions in one place. The tool shells out to native CLIs, supports autonomous goals with full access, and runs on macOS, Linux, and Windows, with mobile and Home Assistant integration. Your AI agents, in concert. Podiom is an open-source, local-first workspace for Claude Code and OpenAI Codex. Give each agent a name and durable context, then chat, schedule work, or hand it a goal to pursue over time. Podiom keeps the sessions, projects, tasks, progress, and decisions together while the native CLIs do the work. A goal keeps the outcome, progress, and evidence in one place. Claude Code and Codex are easy to start in a terminal. They are harder to manage once you have several sessions, projects, and unattended jobs running at the same time. Context ends up spread across terminal tabs, and you spend time reconstructing what happened and what an agent should do next. Podiom puts those agents in one local workspace. Sessions stay available after the backing CLI session changes. Projects carry shared context. Roadmap tasks and schedules keep work moving. You can see which agent did the work, open the session that produced it, and continue from there. Podiom does not replace the provider runtime. It shells out to the native claude and codex CLIs and uses their models, MCP servers, tools, skills, and authentication. A task tells an agent what to do once. A goal /Podiom/Podiom/blob/master/docs/goals.md gives one lead agent an outcome to own over days or weeks. Describe what you want, define what "done" means, add optional metrics, and choose how often the agent should review its progress. - The lead agent turns the outcome into roadmap tasks and schedules, delegates work when useful, and changes the plan as it learns. - Each review records progress, evidence, metric changes, and the next step the agent intends to take. - Missing capabilities become structured access requests. Work that only you can do becomes an action item instead of disappearing into a chat transcript. - Every run and tool call stays attached to the goal. The agent can propose completion, but only you can mark the goal done. Important Goals are deliberately autonomous. The lead agent and every linked task or schedule run with full access and without per-action approval prompts. Podiom records the resulting tool activity on the goal timeline. Read the Goals guide /Podiom/Podiom/blob/master/docs/goals.md goals-run-in-yolo-mode before assigning a goal. Each agent has a name, workspace, identity, provider, model, profile, permission mode, and fallback defaults. Claude and Codex agents can work side by side without flattening their provider-specific behavior. Named agents keep their own identity, workspace, and runtime defaults. Podiom stores a canonical history for every session. If you switch provider or profile, it can replay that history onto a fresh backing CLI session. Scheduled runs and roadmap tasks create normal sessions too, so unattended work is not a dead-end log. Open the exact session behind a task, schedule, or goal run and continue the conversation. Projects give every agent the same source context and standing instructions. The Roadmap holds assignable tasks, while the embedded scheduler handles recurring routines and timed pickups. Agent-created work keeps its provenance, so you can trace it back to the decision that created it. The examples cookbook /Podiom/Podiom/blob/master/examples/README.md has copyable schedules, a runnable goal-creation script, a shared project ledger entry, agent defaults, and a no-credential MCP server example. Use it after onboarding when you want a real first workflow instead of a blank config file. Podiom stores its state under one configurable local root and ships the web UI inside podiomd . It runs on macOS, Linux, and Windows, and the same core can run as a Home Assistant add-on or in a container. Provider CLIs keep their native authentication and policy controls. The native iOS and Android apps can connect to standalone daemons or an opt-in, API-only Home Assistant LAN endpoint; see the mobile /Podiom/Podiom/blob/master/docs/mobile.md and Home Assistant /Podiom/Podiom/blob/master/docs/home-assistant.md guides. - Developers already using Claude Code or OpenAI Codex locally. - Builders who want persistent, reviewable agent work instead of disposable terminal sessions. - Maintainers who need recurring jobs, project context, and an audit trail in one lightweight workspace. - People who prefer local state and native tools over moving their workflow into another hosted agent runtime. Install the latest release on macOS or Linux: Homebrew users can install Podiom with brew install Podiom/podiom/podiom . curl -fsSL https://github.com/Podiom/Podiom/releases/latest/download/install.sh | bash On Windows PowerShell: irm https://github.com/Podiom/Podiom/releases/latest/download/install.ps1 | iex The installer downloads the matching release binary, verifies its checksum, can set up user-level autostart, and launches podiom onboard to check Claude and Codex and create your first agent. Linux releases are distro-neutral static binaries. Open http://127.0.0.1:8787 http://127.0.0.1:8787 after onboarding. Updates are available from the CLI or web UI: podiom update check podiom update apply --yes Prerequisites: Go 1.26+ and Node 20+. Build the Vite web UI and both binaries into bin/ with a version stamp. make build Run the daemon in the foreground. It scaffolds ~/.podiom on first run. ./bin/podiomd In another shell, check that it is live. ./bin/podiom status Open http://127.0.0.1:8787 http://127.0.0.1:8787 for the web UI. For frontend development with hot reload, run npm run dev in web/ . Vite proxies API and WebSocket traffic to a running podiomd . Every commit to master publishes a GitHub Release using the monotonic v0.1.