rajit

why would someone use this tool over Codex? Can't I use Codex with any model?

It is possible to do it (ex: run a background process that analyzes memories across customers and updates the system prompt based on the findings). A specific implementation would depend on your application. Feel free to email me [email protected] if you'd like to talk further.

Yes by using us developers are deferring the harness engineering onto us, and they can spend time writing code for their business logic.

We are closer to infrastructure than a library or framework; we give developers a live agent they can chat with in a single API call.

Good question. There a few differences between our approach and shipping an agent with the Claude agent sdk.

1. Our approach has cron-based or trigger-based automations built-in. Building automations with claude agent sdk requires setting up separate infrastructure.

2. Our approach has self-learning built-in. Building a feature like "dreaming" https://docs.openclaw.ai/concepts/dreaming with claude agent sdk also requires setting up separate infrastructure.

3. Our approach decouples the harness and the compute, which lets developers enforce a stricter security boundary, while claude agent sdk ships with the harness, shell, and filesystem in one process https://platform.claude.com/cookbook/claude-agent-sdk-07-hos....

4. Our approach does not vendor lock developers.

You could pick the latest harness and then switch when another better one rolls out. Our bet is that a developer's time is better spent speaking to their customers than switching harnesses.

Developers with customer-facing chat products are the ideal customer.

If a startup has a specific flow they want the agent to take and their traffic is bursty, then I'd recommend using a framework like Mastra and deploying onto a sandbox.

For long-running always on agents where it's important to learn the users preferences overtime, our approach is the highest ROI.

> what are the cost and security implications?

Cost is the token usage and container uptime.

> One Docker container per-customer sounds like it would be really expensive.

The advantage is per-user memory and self-learning. For context, Claude Managed Agents uses one sandbox per session: https://platform.claude.com/docs/en/managed-agents/environme....

> Are they started on-demand, or run 24/7?

24/7 (best for customer-facing chat products).

> What keeps users from using the agents for general purpose tasks, protects against prompt-injection, etc?

Users define their agent with a system prompt, tool definitions, and skills (which separate a media generation agent from a people search agent). We use Openrouter which has a prompt injection detection feature: https://openrouter.ai/docs/guides/features/guardrails/prompt....

The most valuable pieces of information an AI agent startup can gather is access to their customer's proprietary data and knowledge of their customers preferences (memory + self-learning).

Even as the cost of writing code goes to zero, those two pieces of information are non-commodities.

Thanks for the feedback. The main idea is that today to built a best-in-class agent, developers build the agent loop, session management, tools, memory, skills, automations (cron + trigger-based), sandboxed deployment, and self-learning.

By providing Hermes with a system prompt, custom tools, and skills, developers get the agent loop, session management, automations, sandboxed deployment, and self-learning for free.

when will the graph memory layer be available?

Yes, email me at [email protected], and I can connect you. Their pricing page is here https://fal.ai/pricing.

Prism and Higgsfield are both similar in that we bring many AI models into one place. Higgsfield is focused on a number of different use cases - storyboarding, ai filmmaking, and visual effects - while Prism is hyper-focused on short form video.

You can give Prism a try at https://prismvideos.com - I'm excited to hear your feedback.

We support Kling 3.0 on our platform, similar to Higgsfield. You can see some presets here: https://prismvideos.com/workspace/templates.

This is a great point, and I agree with you. If a weight loss supplement brand were to use an AI influencer to market their product, it does raise questions about whether their supplement does in fact work on real people.

Nevertheless, things are trending more in this direction, and AI influencers will soon become the norm. Brands should be required to disclose when their marketing is AI.

It's worth mentioning that AI videos on Prism (and on any platform) do not have to be purely prompt to creative. For example, a brand designer can take an existing creative for a billboard for example and then use AI to generate images of this creative at a train station, in the Louvre, at a bus stop etc (without actually going there and shooting images).

This is a great point. It is challenging to know which models are good at what.

We've found that Seedance is good at photorealitic faces, Kling is fantastic at generating audio (highest quality model in terms of syncing character's face to the words they say imo), and Sora is great at UGC.

We access models through Fal (https://fal.ai). We offered day 0 support for Kling 3.0 and launch models on our platform the day they are live.

Would be curious to see your script.

How do you identify "wrong tool" invocations (how is the "wrong tool" defined)?

We spoke to a number of browser agent companies who said deterministic RPA with an AI fallback was their "secret" :)