OpenClaw's Control UI Side chat picked up an important compatibility fix this morning. PR #157474, "fix: Side chat cannot answer on claude-cli installs without an API key," lets Side chat answer through a tool-free, one-shot CLI process when a subscription-backed claude-cli setup does not have a direct provider API key.
The PR closes issue #141483 and is labeled P1, with compatibility and auth-provider merge-risk tags. In plain terms, it handles a common user setup: Claude works through the CLI, but OpenClaw's embedded provider path reports missing-provider-auth because there is no separate API key.
What Users Should Notice
Before this change, Side chat could fail with a missing provider authentication error even though the local Claude CLI was available. After the fix, OpenClaw can route the Side chat question through the CLI in the specific missing-auth case.
The PR is careful about the fallback:
- It only uses the CLI path when credential resolution reports
missing-provider-auth. - Other credential failures are not hidden.
- CLI Side chat runs as a one-shot process.
- Read-only direct-provider installs keep their existing tool and image support.
- CLI image questions return the existing
image-input-unsupportederror.
That means this is not a blanket replacement for provider authentication. It is a narrow bridge for users whose usable model access is already represented by the local CLI.
Why It Matters
Side chat is a small interface with an outsized job. It lets a user ask a question about the current session without turning the main agent run into a detour. If that assistant path fails because the account is available through claude-cli but not through a direct API credential, the feature feels broken even though the model is reachable.
This fix makes Side chat better match how people actually run local AI tooling. A subscription-backed CLI install is a real working configuration, and OpenClaw now treats it that way for this narrow use case.
Guardrails Stay In Place
The implementation still carries the selected logical provider and model into CLI admission, so restricted operators can be checked against the intended route. The candidate proof captured the CLI arguments, including empty tools, disallowed MCP tools, strict MCP config, no session persistence, and a one-turn limit.
That matters because a Side chat fallback should not quietly become a broader tool execution path. The merged behavior is specifically a tool-free question path.
Tested With Synthetic Boundaries
The PR evidence used a real built Gateway and Control UI with isolated state and non-default loopback ports. The external Claude executable and OpenAI-compatible provider were scripted boundaries, while Gateway routing, process launch, session snapshot construction, RPC transport, tool execution, image delivery, and UI rendering were real.
Focused ask and RPC suites passed, and the proof retained direct API-key behavior for installations that already have normal provider credentials.
The Takeaway
PR #157474 is a practical quality-of-life improvement for OpenClaw users who rely on Claude through the CLI. Side chat now has a clearer route through that setup without weakening the direct-provider path or pretending unsupported image input works through the CLI fallback.