Files
PaperClipAI/docs/adapters/overview.md
T
Devin FoleyandPaperclip c48feee190 Improve live agent feedback during sandboxed runs (#8915)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - A core part of that experience is watching active agent runs without
dropping into raw logs first
> - Local and sandbox-backed adapters already record useful run output,
progress, and tool activity
> - But active issue threads could sit visually stale while the agent
was syncing workspaces, tailing sandbox output, or emitting incremental
tool-call updates
> - Operators need timely, human-readable progress while preserving the
raw transcript underneath
> - This pull request streams sandbox run-log progress into runtime
status, keeps visible issue threads refreshed, and folds repeated ACPX
tool updates into stable transcript cards
> - The benefit is that long-running agent work becomes easier to
supervise without changing the task/comment control-plane model

## Linked Issues or Issue Description

No public GitHub issue exists for this exact change.

Problem/motivation:

- During long-running sandboxed agent work, the issue UI can appear idle
even though the agent is actively syncing, running tools, or producing
incremental output.
- Operators need realtime feedback at the issue-thread layer, not only
after opening raw logs or waiting for the final heartbeat result.
- Related public context: #1808 previously added live-run status dots to
Projects; #4362 touches heartbeat wakeup behavior but is not a duplicate
of this runtime/UI feedback change.

## What Changed

- Added sandbox run-log streaming support and defaulted sandbox-capable
local adapters into the richer live-feedback path.
- Surfaced environment/sandbox sync progress through heartbeat runtime
status with bounded, redacted snippets.
- Added live issue-thread cache patching so visible active runs update
as progress events arrive.
- Folded repeated ACPX `tool_call` updates into one transcript card
instead of stacking duplicate cards.
- Updated adapter docs and added focused regression coverage for sandbox
log streaming, runtime status, ACPX parsing, live updates, transcript
rendering, and issue chat messages.

## Verification

- `pnpm install --frozen-lockfile`
- `pnpm exec vitest run ui/src/context/LiveUpdatesProvider.test.ts`
- `pnpm exec vitest run
server/src/services/heartbeat-run-runtime-status.test.ts
server/src/__tests__/heartbeat-runtime-state.test.ts
ui/src/context/LiveUpdatesProvider.test.ts`
- `pnpm exec vitest run
packages/adapter-utils/src/execution-target-sandbox.test.ts
packages/adapter-utils/src/sandbox-managed-runtime.test.ts
server/src/services/heartbeat-run-runtime-status.test.ts
server/src/__tests__/agent-live-run-routes.test.ts
server/src/__tests__/heartbeat-runtime-state.test.ts
packages/adapters/acpx-local/src/ui/parse-stdout.test.ts
ui/src/context/LiveUpdatesProvider.test.ts
ui/src/components/transcript/RunTranscriptView.test.tsx
ui/src/lib/issue-chat-messages.test.ts
ui/src/components/IssueChatThread.test.tsx`
- GitHub PR workflow on head `8397953e7b41ccd42e5d9457ee7e4dfb996e4ec5`:
`verify`, build, typecheck/release-registry, e2e, general shards,
serialized server shards, and canary dry run passed.
- Greptile Review on head `8397953e7b41ccd42e5d9457ee7e4dfb996e4ec5`:
Confidence Score 5/5, no unresolved review threads.

## Risks

- Live issue-thread cache patching could miss an edge case for a route
shape not covered by tests.
- Surfacing active-run snippets needs continued care around redaction;
this PR keeps snippets bounded and adds redaction-focused coverage.
- More frequent active-run UI refreshes could expose performance issues
on very large issue threads, though updates are scoped to visible
run/query caches.

## Model Used

OpenAI GPT-5 via Codex, operating as a tool-enabled coding agent with
shell, git, and repository-editing capabilities. Context window size is
not exposed in this runtime.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-07-02 22:21:56 -07:00

6.9 KiB

title, summary
title summary
Adapters Overview What adapters are and how they connect agents to Paperclip

Adapters are the bridge between Paperclip's orchestration layer and agent runtimes. Each adapter knows how to invoke a specific type of AI agent and capture its results.

How Adapters Work

When a heartbeat fires, Paperclip:

  1. Looks up the agent's adapterType and adapterConfig
  2. Calls the adapter's execute() function with the execution context
  3. The adapter spawns or calls the agent runtime
  4. The adapter captures stdout, parses usage/cost data, and returns a structured result

Built-in Adapters

Adapter Type Key Description
Claude Code claude_local Runs Claude Code CLI locally
Codex codex_local Runs OpenAI Codex CLI locally
ACPX Local acpx_local Runs Claude, Codex, or a custom ACP agent through ACPX with live structured event streaming
Gemini CLI gemini_local Runs Gemini CLI locally (experimental — adapter package exists, not yet in stable type enum)
OpenCode opencode_local Runs OpenCode CLI locally (multi-provider provider/model)
Cursor cursor Runs Cursor in background mode
Pi pi_local Runs an embedded Pi agent locally
Hermes hermes_local Runs the local Hermes CLI through @paperclipai/hermes-paperclip-adapter
Hermes Gateway hermes_gateway Calls an already-running Hermes API server through @paperclipai/hermes-paperclip-adapter/gateway
OpenClaw Gateway openclaw_gateway Connects to an OpenClaw gateway endpoint
Process process Executes arbitrary shell commands
HTTP http Sends webhooks to external agents

Hermes local vs gateway

Use hermes_local when Paperclip should start the local hermes CLI on the same host for each heartbeat. Use hermes_gateway when Hermes is already running as an HTTP/SSE API server and Paperclip should call that server instead of spawning a process. Both type keys are stable built-ins.

The unified Hermes package owns both built-in adapters. The older @paperclipai/adapter-hermes-gateway package remains only as a deprecated compatibility shim that re-exports the gateway entrypoints for one release. New plugin overrides should target @paperclipai/hermes-paperclip-adapter and set the desired type key (hermes_local or hermes_gateway).

External (plugin) adapters

These adapters ship as standalone npm packages and are installed via the plugin system:

Adapter Package Type Key Description
Droid @henkey/droid-paperclip-adapter droid_local Runs Factory Droid locally

External Adapters

You can build and distribute adapters as standalone packages — no changes to Paperclip's source code required. External adapters are loaded at startup via the plugin system.

# Install from npm via API
curl -X POST http://localhost:3102/api/adapters \
  -d '{"packageName": "my-paperclip-adapter"}'

# Or link from a local directory
curl -X POST http://localhost:3102/api/adapters \
  -d '{"localPath": "/home/user/my-adapter"}'

See External Adapters for the full guide.

Adapter Architecture

Each adapter is a package with modules consumed by three registries:

my-adapter/
  src/
    index.ts            # Shared metadata (type, label, models)
    server/
      execute.ts        # Core execution logic
      parse.ts          # Output parsing
      test.ts           # Environment diagnostics
    ui-parser.ts        # Self-contained UI transcript parser (for external adapters)
    cli/
      format-event.ts   # Terminal output for `paperclipai run --watch`
Registry What it does Source
Server Executes agents, captures results createServerAdapter() from package root
UI Renders run transcripts, provides config forms ui-parser.js (dynamic) or static import (built-in)
CLI Formats terminal output for live watching Static import

Choosing an Adapter

  • Need a coding agent? Use claude_local, codex_local, acpx_local, opencode_local, hermes_local, or install droid_local as an external plugin
  • Need the richest live run feedback (especially for sandbox workers)? Use acpx_local — see Feedback granularity
  • Need Hermes on another host or already running as a service? Use hermes_gateway
  • Need to run a script or command? Use process
  • Need to call a custom external service? Use http
  • Need something custom? Create your own adapter or build an external adapter plugin

Feedback Granularity

Adapter choice determines how much structured, live detail a run's transcript can show while the agent is still working. Every adapter's stdout is streamed to the run log and rendered live in the UI — including runs on sandbox execution targets, whose logs are tailed and delivered incrementally — but the granularity of what you see depends on the event stream the adapter emits.

Rough tiers, richest first:

  1. acpx_local — full structured event stream. ACPX emits a JSONL event per meaningful runtime moment: acpx.session (agent, mode, session identity), acpx.status (progress text plus context-window usage), acpx.text_delta (assistant/thinking token deltas), acpx.tool_call (tool title, call id, and status updates as the call progresses), acpx.result (stop reason summary), and acpx.error (code, message, retryability). The transcript renders these as live-updating message, thinking, tool, and status blocks, and repeated acpx.tool_call status updates fold into a single tool card instead of stacking duplicates.
  2. CLI wrappers (claude_local, codex_local, cursor, opencode_local, …). These parse each CLI's own streaming JSON output. You get assistant text, tool calls/results, and a final usage/cost summary, but granularity is limited to what the CLI prints — some emit tool progress, others only call/finish pairs.
  3. Generic adapters (process, http). Plain stdout/stderr lines with no structured transcript — you see raw output only.

Recommendation: for sandbox workers, prefer acpx_local. Sandbox run logs are streamed live, so the richer the event stream, the more useful the live transcript and status line are while a remote run is in flight. ACPX's status events (including context usage) and incremental tool-call updates give the closest thing to watching the agent work locally.

UI Parser Contract

External adapters can ship a self-contained UI parser that tells the Paperclip web UI how to render their stdout. Without it, the UI uses a generic shell parser. See the UI Parser Contract for details.