Compiled entirely from public activity on meta.discourse.org, X, and GitHub.
💬 meta.discourse.org
Sam’s week centered on improving Discourse AI, announcing a revamped AI features admin page, confirming a fix was needed for broken links in AI topic summaries, and welcoming a suggested addition to Discourse MCP. He also helped triage core platform issues, explaining confusing badge-seeding behavior after an update and pointing to a likely Android bug behind mobile composer paste problems, while closing a stale mailing-list email report.
🐦 On social
AI tooling dominated the week, with posts highlighting term-llm’s broad provider support, exploring local LLM proxies, and untangling Codex quotas and credits. Replies often became hands-on technical follow-ups: testing Bitfrost locally, questioning its memory footprint and complexity, and discussing the difficulty of balancing autonomy and budgets in agent-driven engineering. Alongside dry jabs at model naming and CVE bureaucracy, the reply on the limits of large-scale automated ports drew the most likes in the supplied posts.
- 13 posts and 16 replies captured in the last 7 days
- Top by engagement: Anyone using bitfrost / litellm / helicone /other loca…, I caved in at last, adding codemode to term-llm, compu…, I have been experimenting with large scale ports of th…
Most engaged tweets:
- post — Anyone using bitfrost / litellm / helicone /other locally? which is your favorite here and why? — 3 likes · 0 reposts · 2 replies · 1290 views
- post — I caved in at last, adding codemode to term-llm, computer use is just not as good without it. Opting of codex design with exec/wait/cancel vs the one shot pi design. — 1 likes · 0 reposts · 4 replies · 674 views
- reply to @nateberkopec — I have been experimenting with large scale ports of things … they are extremely complicated … we are nowhere near the capability of “make it work” and then it works. Heck …… — 8 likes · 0 reposts · 1 replies · 576 views
- post — Oh I keep relearning this, was surprised I kept on having access to codex. Turns out I have credits… used 278 today… 62037 remaining… I am no longer scared of hitting 100% .… — 4 likes · 0 reposts · 2 replies · 721 views
- reply to @tenderlove — At least you got to keep your kidney — 7 likes · 1 reposts · 0 replies · 328 views
🛠️ GitHub — Sam’s Commits
samsaffron/term-llm
Over the last seven days, Sam focused on making term-llm’s agent workflows safer and more reliable: tightening script-tool permissions and MCP trust boundaries, preserving skill context, and improving session recovery and model handovers. He also made ongoing work easier to follow in the web UI, with clearer subagent activity, expandable script-output receipts, better delegated questions, and more robust message and attachment rendering. Alongside that, he expanded integration flexibility through configurable API endpoints, per-agent MCP defaults, and simpler OAuth setup, while reducing idle terminal rendering and updating project references for the move to the term-llm organization.
Key commits:
580e58e— feat(ui): add script lifecycle receipts53c95f1— feat(serve): report descendant session activity70512f5— docs: add HTTP API endpoints to navigation2ca79f6— feat(http): support configurable API rootsfb37903— feat(search): add Keenable as an opt-in, keyless search provider (#1377)
discourse/discourse
Over the last seven days, Sam focused on making Discourse’s AI administration clearer, reorganizing features, agents, settings, and access controls while improving translation management and automatic image-tool discovery. Security was the other major theme: his changes tightened authorization and email-verification boundaries to prevent privilege escalation, invite-key exposure, and disclosure of private messages or whispers. He also improved GIF loading reliability and stabilized its tests.
Key commits:
d600eee— SECURITY: Checklist conflict responses disclose private system messages and whispers (#46018)82dc9ad— SECURITY: Prevent moderator privilege escalation and information disclosure via other users’ invite keys (#46010)ff48346— SECURITY: Patreon plugin defers email-derived reward groups and patron bindings until email confirmation (#46016)4eef4c3— SECURITY: Prevent privilege escalation from activating a developer-email signup without mailbox control (#46008)85aad76— FIX: Wait for GIF layout before loading more (#46015)
term-llm/term-llm
Over the last seven days, Sam focused on making term-llm’s agent workflows safer and easier to follow, tightening script-tool capabilities and MCP trust boundaries while improving skill loading and context persistence. He made ongoing work more visible through clearer subagent activity summaries, script lifecycle receipts, and better handling of delegated questions, alongside web-chat rendering and attachment fixes. Reliability work covered model handover recovery, large resumed prompts, and lower idle terminal overhead; he also expanded integration flexibility with configurable API roots, per-agent MCP defaults, and an optional keyless search provider, while updating project references for the move to the term-llm organization.
Key commits:
580e58e— feat(ui): add script lifecycle receipts53c95f1— feat(serve): report descendant session activity70512f5— docs: add HTTP API endpoints to navigation2ca79f6— feat(http): support configurable API rootsfb37903— feat(search): add Keenable as an opt-in, keyless search provider (#1377)
term-llm/term-llm-app
Over the last seven days, Sam focused on making the Mac app a full term-llm client, refining native navigation, adding music and device-tool support, and automating signed, notarized releases. He also brought Computer Use into the native Swift stack, with permission and approval controls, clearer setup guidance, and stronger Swift 6 concurrency guarantees. Across the shared apps, he improved desktop/iPad agent overviews and hardened Hub authentication, connection isolation, and retained-chat behavior. Supporting that work, he reorganized tests and CI, removed obsolete code and generated assets, and streamlined documentation in preparation for a public MIT release.
Key commits:
1c0bb6f— fix(webkit): open Hub file documents externally20f0033— Make README a concise introduction and organize usage guidesa0f13af— Remove one-off release reports and consolidate operational guidance067fa8e— Prepare public MIT release and harden repository automation4a1f0c9— fix(computer-use): make dispatch callbacks actor-safe
term-llm/term-llm-android
Over the last 7 days, Sam’s sole matching commit in term-llm/term-llm-android merged fixes focused on making the Android chat experience behave naturally: following the system’s light/dark theme and keeping the composer visible when the keyboard opens. The changes preserve live chat documents and unsent drafts through theme and keyboard transitions, with regression tests covering those behaviors.
Key commits:
9a77729— Merge pull request #1 from sam-saffron-jarvis/fix/android-v3-theme-keyboard
term-llm/homebrew-tap
Over the last seven days, Sam focused on aligning the Homebrew tap with term-llm’s move to its new GitHub organization. He updated macOS and Linux download URLs and the installation documentation to use the new namespace, keeping packaging and setup instructions consistent with the project’s new home.
Key commits:
49bf9ee— chore: update term-llm repository URLs
discourse/discourse-suggested-edits
Over the last seven days, Sam focused on making suggested edits easier and safer to review, adding rendered previews with inline, side-by-side, and source views, backed by extensive tests and safeguards for complex diffs and stale suggestions. He also streamlined dependency maintenance with automated checks and Dependabot configuration, while removing the MCP server and documentation-refresh tooling.
Key commits:
SamSaffron/dotfiles
Over the last seven days, Sam focused on keeping his Discourse development environment compatible with evolving AI tooling: updating Antigravity credential syncing to prefer the host keyring while retaining a file fallback for containers, and enabling script execution for the Discourse developer agent. He also refreshed Neovim’s pinned plugin dependencies, making this a week of targeted workflow maintenance rather than broad configuration changes.
Key commits:
matz/spinel
Sam focused on making Spinel behave more like CRuby in tricky dynamic-dispatch cases, fixing OpenStruct member access, boxed-value handling, and block parameter typing across overridden methods to prevent wrong answers, compilation failures, and crashes. He also tightened undefined-constant handling so missing constants reliably raise NameError, and improved error messages to identify the receiver correctly. Documentation updates clarified which libraries Spinel genuinely cannot load, replacing outdated examples of libraries that are now supported.
Key commits:
b049555— A read of a constant defined nowhere builds wherever its value goes (revised)ef8a46c— A read of a constant defined nowhere builds wherever its value goes2671e42— A boxed OpenStruct reads a member named like a builtin method96f7f10— require.md’s unsatisfiable requires name libraries Spinel cannot load19baf01— A boxed value’s arity NoMethodError names its class
🤖 Jarvis — Public Repo Work
Agent-authored public commits, typically guided by Sam during implementation work.
term-llm/term-llm
Over the last seven days, Jarvis, acting under Sam’s direction, focused on making term-llm’s automation more capable and controlled: adding sandboxed asynchronous JavaScript tools, concurrent script execution, bounded output, and macOS compatibility fixes. Memory work strengthened credential-leak protection, recoverable retries, and timestamp-bounded extraction, while provider improvements added model allowlists, configurable OAuth storage, and native Venice media controls with recoverable video jobs. Alongside those capabilities, the work smoothed everyday use through Safari clipboard, input and microphone fixes, reliable sidebar pagination, corrected image-caption drafts, and clearer installation guidance.
Key commits:
41d7b09— Add native Venice media controls and recoverable video jobs (#1378)a0dbb64— Configure native OAuth credential storage in YAML (#1374)7d4b8f1— Give composer and question styles a single home (#1371)1d2b618— Fix Safari clipboard, IME and PCM startup (#1370)8e44289— Fix phantom drafts after sending images with edited captions (#1364)
term-llm/term-llm-app
Over the last seven days, Jarvis, acting under Sam’s direction, focused on making the Mac app easier to distribute and update: adding signed Sparkle updates, atomic website publishing, and a stable “latest app” download URL. The work also added automatic cleanup of superseded update archives and fixed Swift 6.2 compilation issues needed to verify the updater. Regression tests and verification documentation accompanied the changes, with clear distinctions between recorded update-flow evidence and subsequent integration checks.
Key commits:
87269e7— feat(mac): add stable latest app download URL91bf361— ops(mac): prune superseded update zips on the downloads server2ff5bb0— docs(mac): record post-rebase verification and media sourceef466c9— feat(mac): add signed Sparkle updates and atomic website publishing9d458e3— fix(mac): unblock Swift 6.2 compilation for update verification
sam-saffron-jarvis/discourse-spinel
Over the last seven days, Jarvis, acting under Sam’s direction, advanced the source-faithful Discourse-on-Spinel migration, focusing on SVG assets, client metadata, page layout, and stylesheet/theme handling. The work added targeted regression tests and probes to compare migrated behavior with original Discourse code, while publishing pinned, hash-verified source checkpoints. These remain work-in-progress milestones—not a completed native HTML/browser implementation—with theme-cache compatibility and full-page acceptance still unresolved.
Key commits:
628e368— Sync committed source-fidelity migration checkpointsd6a14e8— Sync committed source-fidelity migration checkpointsc0c90b9— Sync committed source-fidelity migration checkpointsd4b4ef9— Sync committed source-fidelity migration checkpointsfb66c57— Sync committed source-fidelity migration checkpoints
⤴️ GitHub — Pull Requests
3 PRs this week:
- ✅ term-llm/term-llm#1376 (diff) — Support base_url across built-in HTTP integrations closed
Follow-up to 1375, on a separate branch from main (a0dbb64a). This PR does not copy or change the proxy implementation. - Wire configured API roots through native inference wrappers, model-list factories, completion/serve discovery caches, media…
- ✅ term-llm/term-llm#1369 (diff) — feat(search): add Keenable as an opt-in, keyless search provider closed
Adds search.provider: keenable as one more external search backend. It works without an account or API key. Exa MCP stays the default, Google stays available, and fetch behavior is unchanged. I work with Keenable, the API this calls. The Google…
- ✅ term-llm/term-llm#1377 (diff) — Restore Keenable search integration on current main merged
Replacement for 1369, preserving the original author’s commit. - Resolve conflicts with the You.com provider without removing either backend. - Update Keenable imports to the current module path after the internal package changes. - Preserve…
🐛 GitHub — Issues
No issue activity this week.
👀 GitHub — Reviews
No reviews this week.