* feat(ai): add Cloudflare Workers AI as a provider
Cloudflare Workers AI hosts open-weight LLMs (Kimi K2.6, GPT-OSS,
GLM-4.7, Llama 4, Gemma 4, Nemotron 3) on Cloudflare's GPU network with
an OpenAI-compatible endpoint. Reuses the openai-completions API
protocol; the per-account URL contains a {CLOUDFLARE_ACCOUNT_ID}
placeholder resolved at request time by a small helper.
Pi automatically sets x-session-affinity for prefix caching:
https://developers.cloudflare.com/workers-ai/features/prompt-caching/
Auth: CLOUDFLARE_API_KEY (matches pi's *_API_KEY convention) +
CLOUDFLARE_ACCOUNT_ID. The User-Agent identifies traffic as
'pi-coding-agent' in Cloudflare analytics.
Verified end-to-end against a real Cloudflare account: 17 e2e tests
pass across stream/empty/tokens/unicode/tool-call-without-result/
total-tokens against @cf/moonshotai/kimi-k2.6.
Cloudflare AI Gateway is a separate, larger change (it requires routing
through provider-specific subpaths with the matching API protocol per
upstream) and will land in a follow-up PR.
* refactor(ai): move Cloudflare User-Agent and session-affinity flag to per-model metadata
Instead of conditionally setting them in openai-completions.ts based on
provider detection, declare them as model-level fields in the catalog
(headers + compat). This is consistent with how the github-copilot and
kimi-coding entries already declare their static headers.
packages/ai/scripts/generate-models.ts: emit headers and compat fields
on each cloudflare-workers-ai entry (CLOUDFLARE_STATIC_HEADERS).
packages/ai/src/providers/openai-completions.ts: drop the
isCloudflareProvider conditional that injected User-Agent and the
isCloudflareWorkersAI override of sendSessionAffinityHeaders.
packages/ai/src/models.generated.ts: re-spliced 8 cloudflare-workers-ai
entries with headers + compat.
Behavior is unchanged - verified via fetch interceptor that User-Agent
and x-session-affinity / session_id / x-client-request-id are still sent
on outbound requests. 5/5 e2e tests pass.
Long local-LLM SSE streams (e.g. vLLM buffering large tool calls) exceeded
undici's default 300s bodyTimeout and aborted with UND_ERR_BODY_TIMEOUT.
retry.provider.timeoutMs cannot lift this cap because it controls the
provider SDK's AbortController deadline, not undici's per-socket idle timer.
Pass { bodyTimeout: 0, headersTimeout: 0 } to EnvHttpProxyAgent. Provider
SDKs still enforce their own timeouts via retry.provider.timeoutMs.
closes#3715
Bun compiled binaries have an empty process.env when running inside
sandbox environments (e.g. nono on Linux/macOS). This broke API key
detection and model discovery because all process.env.* lookups returned
undefined.
- Add restoreSandboxEnv() helper that reads /proc/self/environ when Bun
is detected and process.env is empty, populating process.env before
any other code runs (coding-agent/src/bun/cli.ts entry point)
- Add getProcEnv() fallback in env-api-keys.ts for direct @mariozechner/pi-ai
consumers that may not go through the coding-agent entry point
- Add unit tests for restoreSandboxEnv
Bun bug oven-sh/bun#27802 causes process.env to be empty when running inside a Linux sandbox (Landlock/seccomp).
This breaks npm resolution because PATH is missing, crashing pi on startup with:
Failed to run npm root -g: Executable not found in $PATH: "npm"
Add getSandboxEnv() to read /proc/self/environ as a fallback on Linux when process.env has zero keys.
Pass this env to all spawn calls in package-manager.ts.
Also fix runCommandSync error handling: check result.error in addition to result.status, and include result.error?.message in the error text so spawn failures (e.g. ENOENT) produce useful diagnostics instead of "null".
Fixes#3806
supportsPromptCaching, supportsAdaptiveThinking, supportsThinkingSignature,
and the Claude detection in streamSimpleBedrock/buildAdditionalModelRequestFields
all check model.id for Claude model name patterns. Application inference profile
ARNs are opaque and do not contain the model name, so these checks silently fail.
Fix by also checking model.name (user-controlled via models.json or
registerProvider) as a fallback in all affected functions. Added a shared
isAnthropicClaudeModel helper for the common Claude detection pattern.
Fixes#2925
Co-authored-by: Your Name <you@example.com>
Previously, registerProvider() overwrote the stored config for a provider,
so a models registration followed by a baseUrl/headers-only override lost
the models. After refresh(), only the override-only config was replayed,
causing extension-provided models to disappear.
Added upsertRegisteredProvider() that merges defined fields into the
existing stored config instead of replacing it. Fields absent from the
incoming config are preserved from the stored config.
Also adds regression tests for dynamic provider override persistence
across refresh.
DashScope / Aliyun Qwen (OpenAI-compatible) rejects `tools: []`
with HTTP 400 `"[] is too short - 'tools'"`. Five providers used
a truthy check (`if (context.tools)`) that treated an empty array
as "send tools", so `pi --no-tools` produced `tools: []` in the
request body. Matching the Google provider's pattern, we now
guard on `context.tools.length > 0`:
- openai-completions.ts
- openai-responses.ts
- openai-codex-responses.ts
- azure-openai-responses.ts
- anthropic.ts
The openai-completions fallback that emits `tools: []` when the
conversation has tool history (required by LiteLLM / Anthropic
proxies) is preserved via the existing `else if (hasToolHistory)`
branch.
closes#3649
Co-authored-by: 槐聚 <huaiju@zbyte-inc.com>
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Mario Zechner <badlogicgames@gmail.com>
