2026-10-11 16:37 UTC

Seth Curry releases Abyss, a local-first ACP agent containerization framework with middleware support that provisions per-agent Docker environments, mounts, and secrets while keeping agents isolated from the host by default.

state: seedheat: mediumuncertainty: mediumconvergesscott: highagent-sandboxing agent-harnesses acp-protocolSeth Curryeverforward

What is this?

Abyss is a newly announced local-first framework by Seth Curry (GitHub: everforward) for running ACP (Agent Communication Protocol) agents in isolated Docker containers with middleware support. The Show HN title indicates it provisions per-agent Docker environments, mounts, and secrets while keeping agents isolated from the host by default. No web results were returned (search deadline reached), so this grounding rests solely on the case hypothesis and HN title β€” the framework's exact capabilities, maturity, and adoption are unverified.

Why it matters to Scott

Seth Curry's Abyss independently arrives at the padded-cell / sandboxed-execution architecture Scott's canon treats as load-bearing: per-agent Docker isolation, middleware composition, secrets/mounts at the execution boundary, and a local-first ACP harness. This is a consequential builder shipping the pattern Scott argues for in SiloOS, Code-First Architecture, and the single-tenant appliance model β€” not merely an example of the topic.
ip:framework.siloosip:concept.sandboxed-executionip:concept.runtime-containmentip:framework.code-first-architecturedev:concept.padded-cell-agent-architecturedev:concept.single-tenant-ai-appliancedev:project.openclawradar:abah-offline-firmware-agentradar:agentthread-per-channel-agent-containersradar:agenticos-self-hosted-governance
queries asked of Scott's wikis
  • agent-sandboxing isolation models Docker vs WASM vs process
  • ACP protocol adoption local-first agent harnesses
  • agent-harnesses middleware composition patterns
  • local-inference agent containers secrets management
  • everforward Seth Curry prior work agent tooling

Measured heat

now 0 pts/hpeak 6 pts/hcomments 0/hpeers p14momentum: steady2 platformsage 1178h
points/hour across evidence Β· reading as of 2026-10-12 02:59:37.977291+11:00 Β· deterministic, not a model opinion

How the heat travelled

08-23 14:00⭐ origin echo-reconstructedPrimary artifact is the author's own repo ("A control plane for isolating your LLM agents"). README: "abyss is a system for running your age
Seth Curry (GitHub: SethCurry; HN: everforward; scurry.io) on github (echo) Β· attributed from hn.story.50007023
β€”
10-08 15:29first on hacker news Β· published Β· +1105.5hShow HN: Abyss – ACP Agent Isolation with Middleware Support
everforward
β€”
10-08 15:29amplified on hacker news πŸ‘‘hn.story.50007023
everforward
peak 1 Β· 0 comments Β· 106% of case engagement
10-08 15:39our radar first saw it Β· +1105.7hdiscovery anchor: hn.story.50007023β€”

Evidence (2) β€” ⭐ canonical anchor

sourceobjectauthorscorecomments
🟧 hnShow HN: Abyss – ACP Agent Isolation with Middleware Support
Retrieved article excerpt

Open article Β· Retrieved 2026-10-08T17:49:34.456845+00:00

[Abyss](https://abyss.scurry.io/)






##### Abyss

- [Docs](https://abyss.scurry.io/docs/guides/what-is-abyss/)
- [Blog](https://abyss.scurry.io/blog/)



- [GitHub](https://github.com/SethCurry/abyss)

[Get started](https://abyss.scurry.io/docs/guides/getting-started/)

[Get started](https://abyss.scurry.io/docs/guides/getting-started/)

## Search

Loading search index…

No recent searches

No results for "**Query here**"

- to select
- to navigate
- to close

Search by [FlexSearch](https://github.com/nextapps-de/flexsearch)

# Welcome to Abyss

The easy way to containerize your AI agents.

[Get Started](https://abyss.scurry.io/docs/guides/getting-started/)

Abyss is an orchestrator for running agents in containers.

You configure your editor like a normal ACP agent, and Abyss handles everything Docker for you.

Sample screenshot showing Zed connecting to Abyss

```
docker:
          # This is a pre-built image with Abyss and Pi pre-installed
          image: "ghcr.io/sethcurry/abyss-pi:latest"
        
          # Abyss speaks ACP, so we invoke Pi via pi-acp
          agent_command:
            - pi-acp
        
        host_mounts:
          # This mounts the current directory inside the container.
          # It doesn't take a target path because ACP sends a `cwd`
          # so we need the paths to line up between the host and
          # container
          - source: "./"
        
          # This mounts your .pi directory so that the agent has
          # API keys for your LLM, prior sessions, plugins, etc.
          # 
          # You can put this in the `copy_files` config section
          # which will prevent the LLM from modifying your config files
          # but your sessions won't end up on your host PC so the session
          # files are lost when the container gets deleted.
          - source: "~/.pi"
            destination: "/root/.pi"
```

Works with any ACP client you want!

## How Abyss stacks up

A purpose-built runtime for agents, not a general-purpose container tool you have to bend into one.

|  | Abyss | Docker Compose | Docker Agent | Regular ACP |
| --- | --- | --- | --- | --- |
| Agent-Agnostic | βœ“ | βœ“ | β€” | βœ“ |
| Native ACP Support | βœ“ | β€” | βœ“ | βœ“ |
| Containerization | βœ“ | βœ“ | βœ“ | β€” |
| ACP Middleware Plugins | βœ“ | β€” | β€” | β€” |
| Startup scripts & provisioning | βœ“ | Manual | β€” | β€” |

## See No Evil

Your secrets stay secret. Abyss keeps your agent in the dark until you explicitly shine a light on something β€” nothing is visible by default.

## Hear No Evil

Every filesystem and shell request your agent makes is intercepted and executed inside the container β€” never on your real machine.

## Speak No Evil

Declare your environment once: files to copy, mounts to create, scripts to run. Abyss provisions the whole thing for you, every time.

## Read the docs

Learn more in the [Docs](https://abyss.scurry.io/docs/guides/what-is-abyss/).

Top
everforward10
🟧 echo.github ⭐Primary artifact is the author's own repo ("A control plane for isolating your LLM agents"). README: "abyss is a system for running your ageSeth Curry (GitHub: SethCurry; HN: everforward; scurry.io)β€”β€”

Interpretation history

Decision trace