Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

docs(web-chat): clarify engine ownership and attachment handling

Steven Enamakelc95e232d30c0
diff --git a/crates/openhuman-core/src/web_chat/README.md b/crates/openhuman-core/src/web_chat/README.md
index 39c39f75dc..6e9126be2e 100644
--- a/crates/openhuman-core/src/web_chat/README.md
+++ b/crates/openhuman-core/src/web_chat/README.md
@@ -9,10 +9,10 @@ logic behind the Socket.IO `chat:start` and `chat:cancel` handlers.
Two shared engines do the generic work, and this module is only what OpenHuman
alone can supply:
-| Engine | Owns | Where |
+| Owner | Owns | OpenHuman's host seam |
| --- | --- | --- |
-| TinyAgents' turn runner | The lifecycle of a turn: ingress order, per-thread serialization, the five queue modes, cancellation generations, idle and wall-clock guards, reply-store ordering through a host port, exactly one terminal event | `vendor/tinyagents/crates/tinyagents-harness/src/turn_runner/` |
-| TinyChannels' compiled web projection module | The projection of canonical output events onto typed web events, with per-turn `seq` | `vendor/tinychannels/crates/tinychannels-module/`, called through `tinychannels-bus` |
+| TinyAgents harness | The generic turn runner and run-queue mechanics used by web chat | `agent/queued_turn.rs` supplies the product payload; `runner.rs` and `turn_host.rs` adapt web-chat policy and ports. |
+| TinyChannels module | The projection of canonical output events onto typed web events, with per-turn `seq` | `surface.rs` calls the compiled module through `tinychannels-bus`; the host keeps routing and process-wide event publication. |
Hosted channel domain events are dispatched through this module's `start_chat`
via `channels/bus/subscriber.rs`; those turns share the web-chat runner. Relay
diff --git a/docs/gitbooks/en/features/chat.md b/docs/gitbooks/en/features/chat.md
index 67f9f9b9f0..5abc00ff7f 100644
--- a/docs/gitbooks/en/features/chat.md
+++ b/docs/gitbooks/en/features/chat.md
@@ -16,7 +16,7 @@ Open it from **Chat** in the sidebar, or press `⌘N` for a new conversation.
The input is a rich text field. That is how a `/` becomes a command chip and a popover follows your cursor as you type. Around it you have:
-- **Attach**. Images, PDFs, Office documents and archives are saved in the working folder. The model gets them natively when both the model and the route support that type. Otherwise it gets extracted text, a short readout, or an archive listing.
+- **Attach**. Images, PDFs, Office documents and archives are accepted by the core. Chat messages retain scoped opaque references rather than local paths or inline media bytes. For each turn, the trusted host resolves the references and sends supported content to the model route; unsupported document types use extracted text or an archive listing.
- **Model** and **thinking effort**. Both apply to this conversation only, not globally.
- **Context ring**. It shows how full the context window is for this thread. Click it for a breakdown with the cost of each sub-agent.
- **Voice**. Tap to speak, or hand the turn to the live voice agent. See [Voice](native-tools/voice.md).
@@ -75,8 +75,7 @@ A thread can carry a goal and a to-do list that the agent sets. They show as a p
A long conversation is compacted, not truncated. Compaction seals what came before and starts a new generation that points back to the sealed one. Nothing is deleted on disk, though the model reads only the latest generation. There is no UI for browsing older generations yet.
In Web chat, attachments travel as typed content parts. The core keeps uploaded
-bytes
-behind scoped opaque references in the durable user row; local file paths and
+bytes behind scoped opaque references in the durable user row; local file paths and
base64 data are not used as the persisted message reference. When a turn runs,
the trusted host resolves those references and passes the supported image,
audio, video, or file content to the model route. Voice notes retain typed