Compatibility Reports

Hamr tracks local-model compatibility with explicit reports instead of broad claims. A report records one concrete smoke test against one provider, endpoint, model, and Hamr version.

Use this page as the public format until Hamr grows a generated doctor report. Mark unknown or untested fields as unknown or not tested; do not infer a pass from a similar model or gateway.

Report Fields

Field	Required	Value
hamr_version	Yes	Hamr package version or commit hash used for the test.
tested_at	Yes	Date of the test in YYYY-MM-DD format.
tester	No	Person or automation that ran the smoke test.
provider_gateway	Yes	Gateway name, such as Relay, llama.cpp server, or another OpenAI-compatible server.
provider_version	No	Gateway version when known. Use unknown if unavailable.
base_url	Yes	Endpoint root, with host details redacted if needed.
model	Yes	Exact model ID reported by the gateway.
quantization	No	Quantization label when known, such as IQ3_XXS, Q4_K_M, or unknown.
endpoint_compatibility	Yes	pass, partial, fail, or not tested.
native_tool_calls	Yes	pass, partial, fail, or not tested.
text_tool_calls	Yes	pass, partial, fail, or not tested.
reasoning_leakage	Yes	none observed, sanitized, leaked, or not tested.
max_context_tested	No	Largest configured or observed context used during the smoke test.
diagnostics	Yes	Short pass/fail notes from hamr doctor --full, chat, ask, run, or parser tests.
caveats	Yes	Known limitations, skipped checks, or local setup constraints.

Markdown Template

## Provider / Model

- Hamr version:
- Tested at:
- Tester:
- Provider gateway:
- Provider version:
- Base URL:
- Model:
- Quantization:
- Max context tested:

| Check                  | Result     | Notes |
| ---------------------- | ---------- | ----- |
| Endpoint compatibility | not tested |       |
| Native tool calls      | not tested |       |
| Text-shaped tool calls | not tested |       |
| Reasoning leakage      | not tested |       |

Diagnostics:

- `bun run hamr -- doctor --full`: not run
- `hamr chat` smoke test: not run
- `hamr ask` smoke test: not run
- `hamr run` edit smoke test: not run

Caveats:

-

JSON Shape

{
  "hamr_version": "0.0.22-alpha",
  "tested_at": "2026-05-05",
  "tester": "manual",
  "provider_gateway": "Relay",
  "provider_version": "unknown",
  "base_url": "http://127.0.0.1:1234/v1",
  "model": "Qwen3.6-35B-A3B-UD-IQ3_XXS.gguf",
  "quantization": "IQ3_XXS",
  "endpoint_compatibility": "not tested",
  "native_tool_calls": "not tested",
  "text_tool_calls": "not tested",
  "reasoning_leakage": "not tested",
  "max_context_tested": "unknown",
  "diagnostics": [],
  "caveats": ["Example only; not a verified compatibility result."]
}

Current Matrix

Provider / Gateway	Model	Status	Evidence
Relay	Qwen/Unsloth GGUF models	Assumed target	Product target documented in the PRD; no checked-in smoke report yet.
OpenAI-compatible custom server	Any local model	Assumed target	Config path exists; compatibility depends on endpoint and model behavior.

Assumed target means Hamr is designed for that path, not that the exact provider/model pair has passed a recorded smoke test. Add a dated report when a local endpoint is available and keep caveats visible.