fix(cli): pin pi compat flags so lite pi stops sending store to Anthropic models

[devin-ai-integration]

TLDR

Problem this solves:

• lite pi fails on every Anthropic model with anthropic does not support parameters: ['store']
• pi guesses compat from the base URL and sends store: false to the gateway
• With long cache retention it also sends prompt_cache_key / prompt_cache_retention

How it solves it:

• The provider entry lite pi writes now pins compat: {supportsStore: false, supportsLongCacheRetention: false}
• One gateway URL fronts models with different capabilities, so URL sniffing cannot be right

User Flow

Before: a developer launches lite pi, picks an Anthropic model, and every prompt fails

1. They run lite pi against http://localhost:4000 and pi opens pinned to anthropic-haiku-4-5
2. They type Reply with the single word pong and press Enter
3. pi sends POST http://localhost:4000/v1/chat/completions with "store": false in the body
4. The TUI shows Error: 400: litellm.UnsupportedParamsError: anthropic does not support parameters: ['store'], for model=claude-haiku-4-5

After: the same session answers

1. They run lite pi against http://localhost:4000 and pi opens pinned to anthropic-haiku-4-5
2. They type Reply with the single word pong and press Enter
3. pi sends POST http://localhost:4000/v1/chat/completions without store or the prompt cache fields
4. The TUI shows pong

Relevant issues

Reported by a customer in Pylon #8198

Linear ticket

Resolves LIT-7576

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Shared setup. Real pi 0.85.1 (@earendil-works/pi-coding-agent) driven interactively in tmux, real Anthropic calls. The proxy ran litellm/proxy/dev_config.yaml with the drop_params: True line removed, since the customer runs with the default (drop_params off) and that setting hides the bug

cp litellm/proxy/dev_config.yaml /tmp/pi_repro_config.yaml
sed -i '/drop_params: True/d' /tmp/pi_repro_config.yaml
uv run litellm --config /tmp/pi_repro_config.yaml --detailed_debug --port 4000
export LITELLM_PROXY_API_KEY=sk-1234

Before (67cb34c)

1. PI_CODING_AGENT_DIR=/tmp/pi-before uv run lite --base-url http://localhost:4000 --api-key sk-1234 pi, then type Reply with the single word pong and press Enter
2. TUI pane:

 Reply with the single word pong
 Error: 400: {"message":"litellm.UnsupportedParamsError: anthropic does not
 support parameters: ['store'], for model=claude-haiku-4-5. To drop these, set
 `litellm.drop_params=True` or for proxy:\n\n`litellm_settings:\n drop_params:
 true`\n. \n If you want to use these params dynamically send
 allowed_openai_params=['store'] in your request.. Received Model
 Group=anthropic-haiku-4-5\nAvailable Model Group
 Fallbacks=None","type":"invalid_request_error","param":null,"code":"400"}
                                                   (litellm) anthropic-haiku-4-5

3. Proxy log for that request shows the client body carried 'store': False

After (baa6446)

1. PI_CODING_AGENT_DIR=/tmp/pi-after uv run lite --base-url http://localhost:4000 --api-key sk-1234 pi, then type Reply with the single word pong and press Enter
2. TUI pane:

 Reply with the single word pong
 pong
↑2.1k ↓5 1.0%/200k (auto)                          (litellm) anthropic-haiku-4-5

3. sed -n '/"supportsStore"/,+1p' /tmp/pi-after/models.json

        "supportsStore": false,
        "supportsLongCacheRetention": false

Type

🐛 Bug Fix

Caveats (if any)

Low

• supportsLongCacheRetention: false also drops pi's 24h prompt_cache_retention for OpenAI models behind the proxy, which only Anthropic-incompatible params made necessary
• Users on a released lite pi can add a second provider block with the compat flags until they upgrade

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

Link to Devin session: https://app.devin.ai/sessions/17a4cab15fca4978ab3ae1bb5d05e582
Open in Devin Desktop: https://app.devin.ai/desktop/session/17a4cab15fca4978ab3ae1bb5d05e582?variant=devin

The current tip 31fe5e4537 adds only lint suppression comments to the already-proven implementation, so the live QA at baa64463482ff874959d629c73ce78138f6fc657 remains behaviorally current

---

Note

Low Risk
CLI-only change to generated pi models.json; main caveat is OpenAI models behind the proxy no longer get pi’s long-cache retention hints until compat is tuned per provider.

Overview
lite pi now writes explicit pi provider compat flags so the coding agent stops sending OpenAI-only request fields through the LiteLLM proxy.

provider_block adds compat: { supportsStore: false, supportsLongCacheRetention: false } to the models.json provider entry, because pi infers capabilities from the base URL and a single gateway URL fronts heterogeneous models. That prevents chat completions from including store (which broke Anthropic models with UnsupportedParamsError) and related long-cache prompt fields.

The unit test for the provider block shape is updated to expect the new compat object.

^{Reviewed by Cursor Bugbot for commit 31fe5e4. Bugbot is set up for automated code reviews on this repo. Configure here.}

• baa64463482ff874959d629c73ce78138f6fc657 passes /live-pr-risk

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Greptile Summary

This PR pins pi compatibility flags for the LiteLLM gateway so unsupported store and long-cache parameters are not sent

• Adds provider-level compatibility flags
• Updates the provider block regression test
• Strengthens the lint suppression reason added before this review

Confidence Score: 5/5

The PR appears safe to merge, with the previous suppression-comment concern resolved

The latest change replaces the weak suppression reason with a specific explanation, and no new correctness or repository-rule issues remain

Important Files Changed

Filename	Overview
litellm/proxy/client/cli/commands/pi.py	Adds explicit pi compatibility flags and provides a sufficient reason for the required mutable nested JSON object
tests/test_litellm/proxy/client/cli/test_pi.py	Extends the provider block assertion to cover both compatibility flags

_{Reviews (3): Last reviewed commit: "fix(cli): clarify pi compatibility suppr..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_lite_pi_compat_store (31fe5e4) with litellm_internal_staging (67cb34c)}

[devin-ai-integration]

@greptileai

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[devin-ai-integration]

@greptileai

[mateo-berri]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 31fe5e4. Configure here.}