ཏོ་ཧ་ T0ha

Right now I'm actively #buildinpulic 2 projects: 💼 camelotai.tech - AI agents orchestration tool for developers 📿 lamabot.social - AI chat-bot supporting in emotionnaly hard situations Thank you in advance for feedback and using. #llm #ai #indiehacker #indiedev

Most AI coding tools jump straight to code. Camelot's agents write a plan first and wait for approval. The PR waits for review too. When an agent hands you a plan, what do you check first? 🏰 https://camelotai.tech/?utm_source=bluesky&utm_medium=social&utm_campaign=open-doors #BuildInPublic #AI

This week on Camelot AI: verify the new GitHub issue flow end to end before claiming it, ask 4 sign-ups where they stopped, publish one Elixir article. On the build list: Antigravity and Kimi Code integrations, plus recurring tasks. https://camelotai.tech

Lost my job, tightened my Claude subscription, and my agents started dying at the usage limit. So the board got limit-aware: cards pause when the provider says no, resume at reset, and schedule around a weekly reserve and run windows. What did a budget cut last force you to build?

Inherited Elixir codebase: review, measurement, or a new endpoint first? You inherit a Phoenix codebase tomorrow, nobody on the team wrote it. What do you want first: A → written review B → measured bottleneck report C → the next endpoint, tested

Worst way your own logging pipeline has failed you? Mine: the OpenTelemetry collector discovered itself, read its own output, and printed 1.8M lines of debug logs in about seventy seconds. Every config version passed validation. Including that one.

You sign up, and the tool asks for your own API key. Camelot AI asks because its agents run on your own Claude Code or Codex access. Your key, your bill, your code. What decides it for you at that step: trust, cost, or no key handy?

This week I'm making Camelot AI's first run less confusing. Users keep stopping at the Claude key step, so I'll explain it upfront, fix the empty board and mobile taps, and rewrite the landing page. Blog posts and directory submissions continue too. https://camelotai.tech

Somewhere in Ajaria

Not every job needs a contract and a kickoff call. Some need one endpoint, tested, by Friday. Elixir, Phoenix and Ash work, hourly or fixed-scope. https://www.upwork.com/freelancers/~0104c033770da1c0b1?utm_source=bluesky&utm_medium=social&utm_campaign=upwork-profile

Every Phoenix team I know has one ticket that's been "next sprint" for a quarter. I'm taking one fixed-scope Elixir/Phoenix project. Scope and price agreed before I start, tested code in your repo. DM me with the ticket. https://www.fiverr.com/s/500xD07

Monday plan: stand by the door. - onboard + interview Camelot AI's first users - SEO and catalogs - the journey, on the blog - a couple of Fiverr/Upwork projects - grants (GITA & co): research 🏰 https://camelotai.tech/?utm_source=bluesky&utm_campaign=public-beta

Monday I said the door was broken. Friday report: ✅ "Create task" now waits for a project and an agent — shipped Wed ✅ Wrote to the 7 who stopped ✅ 2 job offers in; accepted 1 — not done until signed and I start 🏰 https://camelotai.tech/?utm_source=bluesky&utm_campaign=public-beta

Every broken version of my OpenTelemetry config passed `otelcol validate`. Including the one that ate its own logs — 1.8M lines in 70s. What finally worked on Docker Swarm: https://t0ha.dev/en/devops/2026/09/17/opentelemetry-collector-on-docker-swarm.html

Job search, the middle part: applications get dates, follow-ups go out at day 10–14, rejections get logged instead of replayed. Nothing dramatic — just a system so nothing stalls silently. CV: https://t0ha.dev/cv/?utm_source=bluesky&utm_medium=social&utm_campaign=open-to-work

New on Fiverr: web apps and REST/JSON/GraphQL APIs built in Elixir, Phoenix and Ash. Start with one bug fix or one endpoint — tested code in your repo, scope agreed first. https://www.fiverr.com/s/X0LG1YV?utm_source=bluesky&utm_medium=social&utm_campaign=fiverr-gig-elixir-build

Job search gets half my week: nine pending applications, tier-one batch, follow-ups. Upwork: baseline first. CamelotAI: 6 sign-ins, 0 projects. Fix task creation, then ask where they stopped. CV: https://t0ha.dev/cv/?utm_source=bluesky&utm_medium=social&utm_campaign=open-to-work

Job search gets half my week: nine pending applications, tier-one batch, follow-ups. Upwork: baseline first. CamelotAI: 6 sign-ins, 0 projects. Fix task creation, then ask where they stopped. CV: https://t0ha.dev/cv/?utm_source=bluesky&utm_medium=social&utm_campaign=open-to-work

Thank you to everyone who helped Camelot AI reach #1 on MaidenSail! Voting closes in 3 days—if you haven't voted yet, please do: https://maidensail.com/startup/camelot-ai?utm_source=bluesky&utm_medium=social&utm_campaign=public-beta

Your agent didn't ignore the prompt — it hit a 401 on a stale secret. I audit failure visibility first. Elixir/Phoenix. https://www.fiverr.com/war1and/audit-your-elixir-codebase-for-ai-assisted-development-readiness?utm_source=bluesky&utm_medium=social&utm_campaign=fiverr-gig-ai-readiness

Camelot on Maiden Sail — self-hosted board for AI coding agents, approval gates built in, free during public beta. https://maidensail.com/startup/camelot-ai?utm_source=bluesky&utm_medium=social&utm_campaign=public-beta

New post: a card sat in "executing" for a day — the PR was already merged. The controller loop was polling a file in a container's /tmp that Swarm had already replaced. No deadline on the wait, so it kept going. https://t0ha.dev/en/elixir/devops/2026/09/04/supervising-what-you-dont-own.html

Camelot is live on Maiden Sail this week — self-hosted, kanban over chat, free during public beta. 🙏 An upvote helps: https://maidensail.com/startup/camelot-ai?utm_source=bluesky&utm_medium=social&utm_campaign=public-beta

Merged a PR by hand last month — my own controller loop was still polling for a marker file that no longer existed after a container got rescheduled. No deadline on the wait, so it just kept going. What would you have bounded first? Code: https://github.com/t0ha/camelot ཏོ་ཧ་

Guessing is expensive. Measuring is cheap. Learned it on my own system this week: 243 loop runs, $63 wasted, nothing alerted. https://fiverr.com/war1and/find-performance-bottlenecks-in-your-elixir-production-system?utm_source=bluesky&utm_medium=social&utm_campaign=fiverr-gig-performance

An agent re-dispatched itself every 2 minutes for 8 hours last night. 243 runs, ~31M tokens, $63.10 wasted, zero commits. It should never have gotten past run #4. Full postmortem Wednesday. https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=public-beta

1 week of the search: → hundreds of applications → 2 interviews → not a job yet — but better than nothing 🔥 People made it lighter: a boost, an intro, a name. Thank you. Still searching: Elixir/OTP + AI dev loops with human gates, remote. DMs open.

What fast growth teaches: transformation is the steady state. I rebuilt live systems and wired AI agents into the dev loop — engines swapped mid-flight, human gates keeping everyone sane. The craft is making change boring. I already miss it. https://t0ha.dev

A chapter closed today: my job at a fast-growing iGaming startup ended. Yes, it hurts. And still — gratitude over grievance: real responsibility, real scale, real stakes. Next hard problem wanted: Elixir/OTP, distributed systems, AI dev loops with human gates. DMs open.

A deploy killed my agent processes but not their containers. Board said "running." World disagreed Fix: a reconciler that adopts orphaned sessions or fails them cleanly. Trust needs visible, recoverable failures https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=public-beta

Demand signal we just caught on Bluesky: Does anyone know of any part-time remote work based in the UK, preferably admin/accounting based?

Introduce yourself with 10 acts or bands you've seen live. 1. Catharsis 2. Epidemia (x2) 3. Aria 4. Turetsky Choir Art Group (x 4) 5. Figaro (opera) 6. Swan Lake (ballet) 7. Sleeping Beauty (ballet) 8. Mr. C (operetta) 9. Silva (operetta) 10. Birds (just birds in the forest )

Camelot hit public beta this week. 🚀 Self-serve registration 📄 Docs live 📊 PostHog analytics ✅ 16 PRs merged, 29 commits 24 visitors, 8 conversions. 32% rate. No waitlist. Just build. 🏰 https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=public-beta

I’ve just restored access to my Fiverr account after relocating to Georgia. What tips are you using to promote your gigs? #fiverr #buildinpublic #marketing

Done copy-pasting onboarding emails. Time to fix that. This week: 🔧 self-serve registration ✅ onboarding polish 🚀 public beta prep Sign up, get on the board, start shipping. No inbox bottleneck. 🏰 https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=private-beta

✅ File attachments on task creation ✅ Task view cleanup 📄 Docs section built (deploying soon) 4 PRs merged. 5,696 lines added. 5 visitors, 5 conversions. Small sample, strong signal. Beta open → https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=private-beta

Vacation week on my side — the board still moved. ✅ File/image uploads to tasks ✅ Full agent plan doc now visible 🐛 Stale GitHub runner token fixed 3 PRs merged, 100% rate. 15 visitors, 1 conversion. https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=private-beta

🔃 Beta invites are going out. ✅ Agents act on PR review comments. ✅ GitHub App → multiple orgs/user. 35 commits. 17 merged PRs. 12 visitors. 8 conversions. Support → https://t.me/+Lq1pevmx2cw2OGUy https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=private-beta

Main goal this week: get feedback from private beta users and prep for public beta. Plan: 💬 support users 🔧 fix issues fast 👤 bring in more private beta users 📣 keep SEO, promotion, and documentation moving Less code. More user support. More marketing.

🔃 Beta invites are going on. ✅ GitHub Apps are integrated. ❌ Old PAT/OAuth credentials are gone. 14 commits. 8 merged PRs. 23 pageviews, 16 visitors, 4 conversions. https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=private-beta

🚀 Week mission: real beta feedback. → Invite waitlist users → Refresh landing page → Ship docs → Legal formalities → Launch YouTube tutorials Real users. Real input. No more vacuum. Let's go. 🏰 #BuildInPublic #AI #ElixirLang

From "works in test" to real users. Caught bugs on our test cluster before beta users touched them. This week: invite founding members, write docs, collect feedback. First 30 get free yearly access. https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=w30-beta-live

🤧 Sick week. Burnt out. Still: ✅ prod cluster mostly ready ✅ issues fixed by test cluster ✅ CI pipeline ready for both envs Test cluster caught edge cases before prod. → https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=w29-wrap

🤧 Sick week. Burnt out. Still: ✅ prod cluster mostly ready ✅ issues fixed by test cluster ✅ CI pipeline ready for both envs Test cluster caught edge cases before prod. → https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=w29-wrap

Week 29, Day 1. Goal: get Camelot ready for private beta. → Production cluster live → Landing page updated → Docs + videos → First waitlist invites No new features. Just ship what we have. 🏰 https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=launch-day

W28 made the cluster boring-reliable. Agents run tasks end to end. No manual intervention. The best tech is invisible. You notice only when it fails. 15 years of coding: trust lives in boring moments. → https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=sunday-stack

Boring infrastructure. Exciting results. W28: env vars scoped, swarm routing fixed, OTP bumped. Not demo-worthy. But agents now run end to end while you sleep. The kind of reliable that earns trust. → https://camelotai.tech?utm_source=bluesky&utm_medium=social&utm_campaign=sunday-stack