Product: Remove Ollama and local-model support #25

Merged

creatidy merged 1 commit from product/remove-ollama-support into main

2026-09-05 12:53:37 +00:00

creatidy commented

2026-09-05 12:48:33 +00:00

(Migrated from github.com)

Closes #24

Product decision

Remove Ollama and all local-model support completely. The initial supported environment is OpenAI subscription capacity plus Z.ai Coding Plan subscription capacity. This supersedes PR #23, which was closed without merging or cherry-picking its Ollama parser fix.

Implementation

Deleted scarcity_router/providers/ollama.py and scarcity_router/providers/ollama_acquisition.py.
Deleted the Ollama parser/acquisition tests and all tests/fixtures/ollama-local/ fixtures.
Removed Ollama registration/imports, status configuration, runtime rendering, CLI options and environment-variable reads.
Simplified status to exactly openai, then zai; provider failure isolation remains.
Removed LocalRuntime, CapacitySnapshot.local_runtime serialization/deserialization, and five local-only diagnostics.
Migrated the normalized capacity contract from schema v1 to schema v2 because the field removal is incompatible. No compatibility framework was added.
Incremented model-policy.json policy_version from 2 to 3 without changing its artifact schema_version.
Updated AGENTS.md, README, architecture, product, security, provider, capacity, selection, competitive, evidence, decision and roadmap documentation.

M1 outcome

M1 remains NOT YET PASS. The exact remaining blocker is: live OpenAI subscription-capacity telemetry must produce usable current quota windows in the owner’s supported environment before status can fully replace routine quota checks. No OpenAI debugging exercise was mixed into this PR, and M2 implementation was not started.

Validation

Baseline on clean main: 470 tests.
This branch: 404 tests; obsolete local tests were deleted.
uv sync --only-dev: pass.
make check: pass; basedpyright: 0 errors, 0 warnings, 0 notes.
CLI help exposes only status --json; live JSON contains exactly OpenAI and Z.ai.
git diff --check origin/main: pass; one commit, no merge commits.
Residual scan: zero unwanted active Ollama/local-model references in product code. Remaining allowed references are the explicit no-support rules in AGENTS.md/README, D-017 and superseded decision history, the schema-v2 removal note, and negative contract tests proving the removed options/fields/diagnostics are rejected.
creatidy/ollama-monitoring was not modified and is not referenced as a future selector input.

Closes #24 ## Product decision Remove Ollama and all local-model support completely. The initial supported environment is OpenAI subscription capacity plus Z.ai Coding Plan subscription capacity. This supersedes PR #23, which was closed without merging or cherry-picking its Ollama parser fix. ## Implementation - Deleted `scarcity_router/providers/ollama.py` and `scarcity_router/providers/ollama_acquisition.py`. - Deleted the Ollama parser/acquisition tests and all `tests/fixtures/ollama-local/` fixtures. - Removed Ollama registration/imports, status configuration, runtime rendering, CLI options and environment-variable reads. - Simplified status to exactly `openai`, then `zai`; provider failure isolation remains. - Removed `LocalRuntime`, `CapacitySnapshot.local_runtime` serialization/deserialization, and five local-only diagnostics. - Migrated the normalized capacity contract from schema v1 to schema v2 because the field removal is incompatible. No compatibility framework was added. - Incremented `model-policy.json` `policy_version` from 2 to 3 without changing its artifact `schema_version`. - Updated `AGENTS.md`, README, architecture, product, security, provider, capacity, selection, competitive, evidence, decision and roadmap documentation. ## M1 outcome M1 remains **NOT YET PASS**. The exact remaining blocker is: live OpenAI subscription-capacity telemetry must produce usable current quota windows in the owner’s supported environment before status can fully replace routine quota checks. No OpenAI debugging exercise was mixed into this PR, and M2 implementation was not started. ## Validation - Baseline on clean `main`: 470 tests. - This branch: 404 tests; obsolete local tests were deleted. - `uv sync --only-dev`: pass. - `make check`: pass; `basedpyright`: 0 errors, 0 warnings, 0 notes. - CLI help exposes only `status --json`; live JSON contains exactly OpenAI and Z.ai. - `git diff --check origin/main`: pass; one commit, no merge commits. - Residual scan: zero unwanted active Ollama/local-model references in product code. Remaining allowed references are the explicit no-support rules in `AGENTS.md`/README, D-017 and superseded decision history, the schema-v2 removal note, and negative contract tests proving the removed options/fields/diagnostics are rejected. - `creatidy/ollama-monitoring` was not modified and is not referenced as a future selector input.