OpenAI's DevDay dominated the week
OpenAI held DevDay 2026 on September 29 in San Francisco and used it to ship more than 20 announcements in one sitting. The headline is GPT-6.1 Sol, a model OpenAI is pitching as landing close to flagship GPT-6 Astra quality at roughly a fifth of the price — aimed squarely at teams that found Astra too expensive to run in production. Alongside it: dots, a new class of always-on agents with their own persistent cloud computer and connected apps (texting support and multi-dot teams are coming later), an Ultrafast inference tier, a Pro 500 plan at $500/month, and a security-focused Codex Security Cloud. OpenAI also expanded its Agents API and rolled out a marketplace letting enterprise customers spend committed OpenAI spend on partner products, starting with Baseten for open-source model access.
It's a lot to digest in one recap, but the throughline is consistent with the rest of this year: OpenAI is optimizing for agents that run unattended for long stretches, and pricing tiers that make that affordable enough to actually deploy.
Product Hunt: agents that remember and agents that act
Product Hunt's week was quieter on single breakout launches and heavier on incremental releases from repeat players — Lovable, n8n, Attio, PostHog, Vapi, Granola, Framer, and Replit all showed up again in this week's top slots, mostly with iteration rather than net-new products. The more interesting signal is in the categories Product Hunt is now tracking as first-class: AI Agents, AI Coding Agents, AI Voice Agents, and Agent Identity are all now dedicated leaderboard categories — a tell for where the maker crowd is spending its time. Coverage of the daily leaderboards through the month also flagged CrawlRaven MCP, Agent Identity, and Ferndesk climbing fast, all riding the same theme: infrastructure for agents to act on a user's behalf safely, rather than another chat wrapper.
GitHub Trending: memory, CLI-native agents, and fleets of coding agents
Four repos worth knowing about from this week's trending page:
- vectorize-io/hindsight (43.4k stars) — "Agent Memory That Learns." A learning memory layer for agents that's climbing fast, positioned against static RAG for anything that needs to remember across sessions.
- HKUDS/CLI-Anything (51k stars) — makes any piece of software agent-native from the command line, backed by a CLI-Hub directory. Part of the broader push to make existing tools callable by agents without custom integrations.
- stablyai/orca (82k stars, YC-backed) — an ADE (agent development environment) for running a fleet of parallel coding agents — Claude Code, Codex, Cursor, OpenCode — under one roof, with desktop, mobile, and remote runtime support. The topic list alone (worktrees, parallel-agents, orchestration) says where multi-agent coding tooling is heading.
- trycua/cua (27.5k stars) — open-source computer-use drivers and cross-OS fleets for training, evaluating, and generating data for computer-use agents. Relevant to anyone building or benchmarking their own "agent that clicks around a screen."
Also worth a mention: debpalash/VoiceStudio (49.6k stars), a fully local, open-source ElevenLabs alternative doing voice cloning, dubbing, and TTS across 646 languages — and cloudflare/security-audit-skill (23.3k stars), Cloudflare's own coding-agent skill for running multi-phase security audits with independently verified, machine-readable findings.
Worth your time
If you're only trying one thing this week: stablyai/orca if you're running more than one coding agent in parallel and tired of juggling terminal tabs, or hindsight if your agent keeps forgetting context between sessions and RAG isn't cutting it. On the OpenAI side, GPT-6.1 Sol is the one to actually test against your current model bill — a near-Astra model at a fifth of the price is the kind of change that's worth re-benchmarking for, not just reading about.