zackham
Joined 17 years ago
Karma
763
Submissions
164
Member since
Jul 18
Recent submissions
Comment · 15 days ago
Like others here I created my own terminal multiplexer as well. Daemon that holds all sessions, seamless binary upgrades so those sessions never get lost, everything recorded so recovery is always possible, thin client mode so laptop has a native app that connects to daemon, and a wasm responsive webui with transcription so I can carry on with sessions from phone. Claude/codex sessions started with a uuid so they can be killed/resumed at will (mine autosleep after 12h inactivity to free resouces). CLI allows agent control of everything for a harness independent implementation of claude's channels. A few friends have forked and adopted it as their daily driver. Similar organization to herdr wrt spaces to organize sessions. Optimized for speed/bandwidth when running in thin client mode.
Comment · 3 months ago
single author, single user software is the one place where there is no rev limiter on ai. i spend the vast majority of my time on the computer steeped in my personal software now. some i publish for others to use, or so friends can point their agent at to adapt/rebuild:
https://stepwise.run/ Portable workflow orchestration for agents and humans
https://github.com/zackham/aloop Embeddable Python agent loop for integrating LLM agents into your projects. Multi-provider (OpenRouter, OpenAI, Anthropic, Google, Groq), built-in tools, hooks, ACP server, and persistent sessions.
https://github.com/zackham/plat An embeddable infinite canvas to build on. Bring your own node types; drag, zoom, edges, auto-layout, and undo come free. Host-agnostic storage, runtime API built for agents.
https://github.com/zackham/seance Shared live workspace for humans and agents — multi-pane terminals, notes, and a control plane for co-working in the open.
deep dive docs for my personal agent system i've been developing since dec, and using constantly: https://vita-docs.ham.xyz/overview/
browser replacement for stuff i control: https://vita-reports.ham.xyz/s/2119130b0b6b
a snapshot in time of the education system i run for my kids: https://vita-reports.ham.xyz/s/e7036ae931a7
if you want to talk about any of this stuff, or have a related project you think i'd appreciate, hit me up [email protected] or @zackham on telegram. abundance is here for personal software. take advantage of it and have fun!
Comment · 3 months ago
all ingested data and all derivations of that data can link to source channel and associated ACL that syncs with the source. then search/reads/etc just join on and respect the ACL. not that complicated to manage for things like slack, meetings, etc.
Comment · 3 months ago
agreed. i built this for a wide variety of background workflows that have been running seamlessly for many weeks now. it's an overbuilt personal project but i think pretty dialed in and very flexible https://stepwise.run/
Comment · 6 months ago
I have a project in this space that I've run many thousands of jobs through. It's solid and full featured. Feel free to connect: https://stepwise.run/
Comment · 6 months ago
if you are looking for a customizable agent harness that provides everything except a tui (cli, acp, embeddable python package) i built and use this for a handful of projects where i needed a vendor-agnostic replacement for claude agent sdk: https://github.com/zackham/aloop
Comment · 6 months ago
The most durable way to reason about agents is to just think about humans. We have thousands of years of prior art on coordinating instantiations of stochastic intelligence. Context, tools, goals, validation, specialization, distribution of labor, coordination... If jobs are bundles of tasks and areas of accountability, maybe it's more effective right now to unbundle and reorganize some of these things. If constraints underperform autonomy, maybe you have to adjust where you operate on that spectrum, and account for it in goal definition and validation. These are not new problems.
Comment · 6 months ago
When I come across a paper/article/insight that connects to one or more things I'm working on, I generate 20-30 minute podcasts that do deep research, connect it back to my work, and target an angle I'm interested in. I can message my agent over Telegram and say "check this out [link] curious who else is working on this problem and if there's anything actionable we can glean from it in relation to [projectX], generate a podcast". It's been surprisingly good, and in my commutes I've found these more compelling to listen to than my normal rotation of stuff.
Here's an earlier version of my podcast generation flow before I built narrate: https://stepwise.run/flows/podcast-deep
Comment · 6 months ago
agreed. the long-running tasks i handle with immediate-background and continually edit message with updates, along with fork-and-reconcile semantics so you don't have to queue messages. also having a supergroup with separate topics for different sessions works really well.
the separate topics approach along with allowing other processes to enqueue messages into the session stream solves state pretty tidily, see "the session as decision-maker" in the linked report
https://stepwise.run - Deterministic scaffolding around nondeterministic work. Observable runs. Human gates. Resume after reboot. No babysitting.
https://github.com/zackham/aloop - An embeddable Python agent loop for multi-agent orchestration, declarative permissions, and editor integration via ACP. Use any LLM from any provider, extend through hooks, embed as a library, drive from the CLI, or expose as an ACP server.
[email protected]