Repository navigation
feature/Bump-pipelex-drop-gateway · L-261009-405281 - #95
Merged
Merged
Conversation
Move the pin from 0.46.4 to 0.83.0, adding the `cli` extra that 0.83.0 says to install wherever pipelex runs in a terminal, and refresh the lock. `pipelex migrate`, run with an empty PIPELEX_HOME so it touched only this project, removed what the Pipelex Gateway releases left, which any pipelex from 0.73 refuses to boot on: the `pipelex_gateway` backend and its model file, the `all_pipelex_gateway` and `example_routing_using_patterns` profiles and the gateway routes of the specific-models example. The active profile moves to `all_enabled_backends`, the profile 0.83.0 ships. It also retired the text and HTML renderings of traced data from `pipelex.toml`. Co-Authored-By: Claude Opus 5.5 <[email protected]>
The committed backend catalogs and model deck dated from the Gateway era:
the Gateway fetched its catalogs remotely, so the local files lacked the
very models the deck named (claude-4.8-opus, gemini-pro-latest), and with
the Gateway gone no enabled backend served any model at all.
The inference files are now the ones `pipelex-agent init` writes for
0.83.0, with their kit manifests, so `pipelex update` finds the deck
current. Only `openai` is switched on, because the deck's default tiers are
OpenAI's GPT-5.6 models, and the routing profile stays the shipped
`all_enabled_backends`, so a backend switched on in a personal
backends_override.toml serves its own models with no routing edit. The
perplexity and legacy pipelex_inference catalogs, which 0.83.0 no longer
ships, are gone.
The storage `uri_format` drops `{primary_id}` and `{secondary_id}`, which
0.83.0 refuses at boot and `pipelex migrate` cannot repair, for the
shipped `{hash}.{extension}`.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Pipelex 0.83.0 checks, when it loads a library, that each step of a sequence reads an input under the concept and multiplicity the earlier step stored it with, and refuses the library otherwise, so every cocode command failed before its run started. Five inputs read a list as a single item: `related_files` and `all_inconsistencies` in doc_proofread, and `git_citations`, `doc_analyses` and `structured_changes` in doc_update. Each now declares the list its producing step stores. Co-Authored-By: Claude Opus 5.5 <[email protected]>
The bump let click rise to 8.5.0, whose generic ParamType leaves typer 0.16's Option and Argument partially unknown under strict pyright, and typer 0.26, the newest pipelex allows, vendors its own click, which no longer types cocode's TyperGroup override. Pipelex 0.83.0 locks exactly this pair. Co-Authored-By: Claude Opus 5.5 <[email protected]>
The root conftest booted pipelex in the `ci` integration mode without inference and through the CLI factory, in the `cli` mode, with it. Both now go through one helper that boots in the `pytest` mode, which the Gateway's terms used to rule out and which a pipelex with usage statistics reads as off, without touching the machine's shared statistics state. A unit test holds the helper to that mode. The deprecated `is_inference_disabled_in_pipelex` gives way to `needs_inference_in_pipelex`. `PIPELEX_USAGE_STATS=0` stays the conftest's switch: 0.83.0 no longer raises GatewayDoNotTrackConflictError, so DO_NOT_TRACK would now be safe, but it would also silence every telemetry destination a developer set up, and the narrow switch is the one pipelex's own test helpers set. CI keeps DO_NOT_TRACK=1, with its comment brought up to date. A boot without inference now loads every enabled backend's models without resolving a key, so `test_dry_run_all_pipes` runs in CI again, and `make agent-test` passes --disable-inference, as `make gha-tests` does, so it no longer needs an OpenAI key to boot. Co-Authored-By: Claude Opus 5.5 <[email protected]>
The README offered a free Discord key read from `PIPELEX_API_KEY` and a bring-your-own-key option the committed routing ignored, the docs and CONTRIBUTING.md asked for an OpenAI key the configuration never used, and .env.example still listed `PIPELEX_GATEWAY_API_KEY`. They now tell the setup pipelex 0.83.0 documents, as it applies to cocode: one OpenAI key in `.env` for the shipped configuration, and for any other provider its key, its backend switched on in a git-ignored backends_override.toml, and the deck's aliases cocode's workflows go through pointed at its models. They also say that the hosted Pipelex API does not apply, since cocode runs its workflows in its own process. The changelog records the bump. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…ofread's inconsistencies The package now carries .pipelex/inference/ and boots Pipelex with CocodeModelManager, which reads it whenever the directory cocode runs from has no inference configuration of its own, so a pip install runs on one OpenAI key instead of the every-backend files Pipelex writes into a fresh home. The provider recipe moves the whole model deck in a personal deck file instead of editing the tracked x_custom_llm_deck.toml, and covers pip installs through pipelex-agent init. doc_proofread batched a pipe returning a list, which stored one list per documentation file where the report step reads one flat list: each file now yields one structured result whose inconsistencies a binding step flattens. Also: make codex-tests runs without inference, the make test help says it boots with inference, .env.example lists every backend's variables, and the test boot no longer dry-runs the library a second time. Co-Authored-By: Claude Opus 5.5 <[email protected]>
Member
Author
|
Review pass, profile 4 (cubic, Codex review and adversarial, code-review), at
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Moves cocode from pipelex 0.46.4 to 0.83.0, which has no Pipelex Gateway:
pipelex migratecleaned the Gateway leftovers out of.pipelex/, the inference catalogs and model deck are now 0.83.0's with only the OpenAI backend switched on, and the bundles declare the list inputs 0.83.0's load-time sequence check requires. The README, docs, CONTRIBUTING.md and .env.example drop the free Discord key andPIPELEX_API_KEYfor oneOPENAI_API_KEY, plus the recipe for switching to another provider. The test suite now boots pipelex in thepytestintegration mode,make agent-testruns without inference, and CI dry-runs every pipe.Closes L-261009-405281
Advances L-261011-a88c00
🤖 Generated with Claude Code
Summary by cubic
Moves cocode from pipelex 0.46.4 to 0.83.0, dropping the Pipelex Gateway so all AI workflows run on your machine with your own provider keys.
OPENAI_API_KEYreplaces the gateway/Discord key everywhere, the shipped configuration is 0.83.0's with only the OpenAI backend enabled, and the package now carries that configuration: apip installboots on one OpenAI key from any directory.Breaking changes
PIPELEX_GATEWAY_API_KEYand the free Discord key are gone; setOPENAI_API_KEYinstead.pipelex_gatewaybackend andall_pipelex_gatewayrouting profile are removed; the active profile is nowall_enabled_backends.Configuration and tooling
.pipelex/inference catalogs, model deck, and routing reset to 0.83.0'spipelex initoutput; storageuri_formatis now{hash}.{extension}.backends_override.tomland point the deck's aliases at its models in a personal deck file — no tracked edits.doc_proofreadanddoc_updatedeclare list inputs as arrays to satisfy 0.83.0's load-time sequence check;doc_proofreadnow flattens each file's inconsistencies into one list the report step reads.pytestintegration mode;make agent-testand CI run without inference, and CI dry-runs every pipe.Written for commit dc73cd8. Summary will update on new commits.