Files
pipelines/docs/chat-backends.md
T

13 KiB

Chat backends

Chats can be created on more than one harness. OpenCode remains the default; Codex is available when its App Server control socket exists, and Claude Code when the claude CLI is installed. The selected backend is stored on the operator-session row and returned as backend on every chat_sessions row, so a restart does not have to guess from the session id.

Codex

The adapter in internal/codex connects to the Codex App Server's Unix WebSocket control socket ($CODEX_HOME/app-server-control/app-server-control.sock, falling back to ~/.codex). It uses the operator's existing Codex login. A new chat issues thread/start; a prompt issues turn/start; history comes from thread/read; questions map from item/tool/requestUserInput to the normal question card and chat_answer returns the JSON-RPC response. Threads created this way are ordinary Codex threads and remain visible to the Codex app.

New threads retain the workspace-write sandbox and inherit Codex's configured approval policy (Pipelines no longer forces never). Command, file-change, network and explicit permission requests appear in the web approval cards. Each JSON-RPC callback has its own identity, including Code Mode subcommands sharing one item. Allow once maps to accept, session grants to acceptForSession, and Reject to decline; explicit permission profiles grant only requested fields. The web labels session scope explicitly, never as a permanent policy change.

Pending questions and approvals survive browser reconnects while the App Server connection lives. Standard MCP form/URL elicitation supports typed answers, Accept/Decline/Cancel and safe external links. Extended OpenAI elicitation is not advertised; unknown request formats return an explicit unsupported RPC error, not a fake empty success or silent auto-approval. App Server socket loss does not provide durable callback recovery; a disconnected transport is reported as an error rather than claiming old requests are still actionable.

Claude Code

The adapter in internal/claude owns one Claude Code CLI process per active session. It starts claude -p with stream-json input and output, partial messages, and --permission-prompt-tool=stdio, then performs the SDK-style initialize handshake before any turn. The operator's existing Claude Code login and configuration are used unchanged; Pipelines does not store a second credential or alter global settings. PIPELINES_CLAUDE may name the executable; otherwise claude is resolved from PATH and common per-user locations.

A new session pins a generated UUID with --session-id. When the process is released (idle, an explicit close, or a host that crashed), the next prompt resumes the same conversation with --resume <id>, so context and the selected model survive. Chat creation is local: the CLI starts on the first prompt or model-catalog read.

Surviving a server restart

A server restart does not stop a Claude Code chat. The installed server runs each CLI inside pipelines-claude-host (cmd/pipelines-claude-host, internal/claudehost), a small helper installed next to pipelines-server. The host starts in its own session and process group, so neither launchd nor the supervisor stops it with the server (the systemd unit uses KillMode=process for the same reason). The host owns the CLI's stdin/stdout. It appends every output line to a per-session spool file with byte offsets and serves a Unix socket. The socket is mode 0600 in a 0700 directory: <user cache dir>/pipelines/claude-hosts/<hash of the database path>/. Two servers of one user never adopt each other's chats. Each host also writes a registry file (<key>.json) there.

The socket speaks one JSON object per line: attach (deliver output from an offset; a newer attach replaces the older connection), in (one stdin frame), ack (output processed up to an offset), status and close. Interrupts, model and mode switches travel as the CLI's own control requests inside in. The host snoops these frames to keep the CLI's initialize response (models, account), the applied model and permission mode, a turn's running state, and the control requests nobody has answered yet (can_use_tool, AskUserQuestion).

On shutdown the server only disconnects (Detach). On start it reads the registry and reattaches to every live host before serving. It replays the spool from its start without broadcasting, which rebuilds the session's stream state, background commands and agents, and pending approvals and questions. Requests already answered are skipped. Output after the last acknowledged offset, written while no server was attached, is then processed live. The turn is shown running if it was running at that offset, watchers get a resync, and the start-up turn probe keeps observing it for notifications. An approval or question asked before the restart can be answered after it. Stop works as usual, and the CLI is never restarted or initialized twice.

The host stops the CLI (EOF, then SIGINT, then SIGKILL to its process group) and removes its files only when one of these happens:

  • The operator closes the chat (chat_close).
  • The server's idle release or live-process cap fires.
  • The CLI exits; the server then gets its stderr.
  • With no server attached, the CLI has been idle for the same 20 minutes, or no server has attached for 24 hours.

If the host itself dies, the server first tries to reconnect once. If that fails, it stops the orphaned CLI's process group (only when that pid's command line still names the session) and removes the host files. A running turn ends with the notice "the Claude Code host stopped unexpectedly; the next prompt resumes this session with --resume". The next prompt then resumes the session.

Limits:

  • A container restart (Docker) still ends every host. So does a reboot or logout. The transcript survives and --resume applies.
  • The spool keeps at most 64 MiB of acknowledged output. Older output is dropped, so background-task state from before that point is not rebuilt after a restart. Unacknowledged output is dropped only beyond 256 MiB; the server then logs the gap and clients reload the transcript.
  • Turn bookkeeping that lived only in server memory restarts fresh: the turn id, and which mid-turn messages are still outstanding. Output that begins a follow-up turn after a reconnect is adopted as a running turn.
  • Without the host binary (PIPELINES_CLAUDE_HOST=off, or not found next to the server or on PATH), the CLI is the server's child as before and a restart resumes it.

Text and reasoning stream from the CLI's partial-message events; completed messages are de-duplicated against the streamed text. Tool calls appear as normal tool cards, Bash commands as the existing background-command rows, and results attach to the call. can_use_tool becomes the normal approval card: Allow once returns the original input, Always allow forwards the CLI's own permission suggestions, and Reject returns a deny. AskUserQuestion becomes the normal question card and chat_answer returns the chosen answers. Interrupt sends the CLI's own interrupt control request.

The model catalog comes from the CLI's initialize response and is switched live with set_model; Pipelines persists the choice for the next turn so a resumed process starts with it. Usage and context totals come from the CLI's get_usage and get_context_usage queries: session cost, processed tokens, context tokens and the model window are reported values, not estimates. Compaction boundaries surface as native compaction activity.

History is read from Claude Code's own session transcript ($CLAUDE_CONFIG_DIR/projects/*/<id>.jsonl or ~/.claude). Entries are merged by message id, tool results attach to their calls, subagent sidechains and local command noise are excluded, and paging is bounded. A session file larger than 64 MiB is reported as exceeding the history bound rather than silently truncated. The live user turn keeps the client's optimistic id when it can be matched to the recorded prompt.

Fresh context (/new)

The composer's /new sends chat_new_session. It keeps the chat and clears the agent's context, like /new or /clear in the harness's own TUI. The server creates a fresh native session on the chat's own harness, in the same directory and with the same title. OpenCode makes a new ses_ session, Codex a new thread and Claude Code a new session UUID. The server then re-keys the operator_sessions row in one transaction. The new row keeps the title, directory, tag, backend, colour, star, creation time, model override and mode. The old row is not deleted: it records replaced_by, is no longer listed, and its native session can still be resumed in a terminal.

The row is re-keyed rather than mapped from a stable id because every other layer is keyed by the native id: rooms, queues, transcripts, shells, exports and resume commands. The reply, and the chat_session_changed push to other connections, carry the new row with previous_id. Each client swaps that row in place, and an open conversation follows it to the fresh, empty transcript. A running turn is refused, so stop it first. A second /new from an id that has already been replaced is refused too.

Claude Code also accepts /clear in -p --input-format stream-json mode, but the CLI then picks the new session id itself. Re-keying keeps one path for all three harnesses and lets the server choose the id.

Chat list activity

Every chat_sessions row carries activity (turn running, running background tasks), and chat_activity pushes each change to every connection (see docs/cli.md). Background tasks are the same shells all three harnesses already report: OpenCode shells, Codex commandExecution items, and Claude Code background Bash and Agent/Task subagents. A chat whose last viewer left stays observed while its tasks run, so their exits are seen. That is bounded by the unattended limits (16 chats, 6 hours) and by the harness: when a Claude Code process exits, its background tasks are reported exited because they ran inside it. After an OpenCode or Codex stream reconnects, the shell set is re-read from the harness.

Web interaction contract

All advertised chat adapters (OpenCode V2, Codex and Claude Code) implement the same server contract for uploads, models, pending approvals, questions, form replies, interrupts, command readback and bounded output. Compile-time assertions and browser regressions protect this contract; support is not satisfied by text streaming alone. Pi's pipeline executor is not advertised as a Pi web-chat adapter, and a harness whose CLI is absent is not presented as a working choice.

Codex uploads use native localImage for verified supported images. Other files are retained intact with an explicit server-local reference; the model's own tools inspect them rather than pretending their contents were already loaded. The saved next-turn model is also respected with uploads. Model sandbox and approval policy remain in force. Claude Code receives images as native image blocks, small text files inline, and larger files by absolute server path; the saved next-turn model is applied before the turn.

Codex's paged history retains tool and reasoning items, including structured MCP errors and Code Mode outputs, instead of discarding everything except text. Unknown item kinds remain inspectable. Tool detail is explicitly truncated at 64 KiB, not silently represented as complete. Running commands and their output use real command items plus read-only thread history; no unsupported terminal endpoint, process control, resumed turn or experimental flag is invented. Output reads are bounded and verify thread ownership. The installed Codex version exposes no API for controlling an agent-owned terminal through the separate, client-owned command/exec process IDs; those IDs are never conflated.

A fresh Codex connection can accept a turn without subscribing to its native thread notifications. The adapter therefore reconciles the latest turn through the read-only paged API and emits an authoritative-history resync when it changes. It does not call thread/resume on open/model selection, so merely watching a chat cannot awaken a queued goal. Native turn IDs protect the queue/UI from stale completion events. A real, isolated tool-free turn passed after replacing the owning connection; no command, permission grant or global setting was involved.

An explicit Send can load a persisted thread if turn/start is refused with the exact application error thread not found: <owned-thread-id>. Only then does the adapter call thread/resume (metadata only, no policy/configuration override) and submit the original message/model/files. Transport failures and unrelated errors are never retried. This loading is not performed by browsing, history reads or model selection. As in Codex's native client, explicitly continuing a thread also resumes that thread's configured execution context.