1. Built with
  2. /
  3. Hermes
Sign in
Hermes
Built with

Hermes

  • FF
    Fesal Fayed@member7804
    Article·2w ago
    Hermes Agent open-source fixes and tool-use fine-tuning

    Contribute to Hermes Agent, an open-source AI-agent platform from Nous Research. Three fixes originating in my pull requests reached the main codebase, repairing an Anthropic crash loop, WhatsApp authorization failures, and malformed conversation context after compression. Fine-tuned an open 20-billion-parameter language model for reliable tool use and released four runnable formats for Apple Silicon, local desktops, Colab, and production inference.

  • FF
    Fesal Fayed@member7804
    Article·2w ago
    Comma.ai video compression challenge

    Built a submission for comma.ai’s public video-compression challenge, which asked entrants to shrink driving-scene data while preserving information used by road and motion models. Reconstructed the full set from one 191,052-byte file and cut the official score 8% below the previous best, ranking first when submitted.

  • FF
    Fesal Fayed@member7804
    Article·2w ago
    Comma Hack 7 and lossy video compression

    I was flown from Miami to San Diego by the comma.ai team for Comma Hack 7 after submitting a solution to their lossy video compression challenge. I trained two custom neural networks: one identified road boundaries and preserved the road pixels needed for autonomous navigation, while the other calculated the appropriate gray pixels to preserve warped motion between frames. I compressed a 37.5 MB dashcam video to 191 KB and reached the top of the leaderboard at the time, improving the score from 0.187 to 0.172 and doubling the progress of the previous 30 submissions combined.

  • JA
    Jordan Ayala@jordan
    Post·1mo ago

    I was tinkering with Telegram and Hermes, but it got awkward as soon as I added a second bot. Discord turned out to be the answer. It has channels and threads, and threads cap how much conversation a bot carries in its context. Telegram has threads too, but they don't work out of the box and the UX is clunky.

    With Discord sorted, I set up a server with two bots, Daedalus and Icarus, named after the AIs in Deus Ex video game. Daedalus runs on GLM and orchestrates the long-horizon work. Icarus runs on something faster and cheaper for quick jobs, which keeps him safely away from the sun 🌞.

    What's your agent stack look like? Share your setup in the comments

  • JA
    Jordan Ayala@jordan
    Article·1mo ago
    Building an Agent Team with Hermes

    Nowadays AI are giving you less freedom. You get one model, one way of working that works fine until you need a fast model for drafting and a precise one for code review, and you're stuck switching between tabs. I started using Hermes Agents because I wanted control over which models run where. ChatGPT's one-size-fits-all approach doesn't work when you need different tools for different jobs. Hermes lets me keep agents in separate profiles, each with its own model and provider. I can switch between Llama for quick drafts and GPT-Sol for precision mid-conversation. It also creates skills by itself, reusable routines that learn from context. When I was revising my resumé, Hermes created a skill that understood my background as an AI Engineer and caught things I'd miss alone. The tradeoff however is simplicity. Hermes has a steeper learning curve than ChatGPT. But once you set up separate profiles for writing, code review, and analysis, it's worth the cost. If you're tired of one-tool-fits-all, try building your own agent team.