mathias/tapir
1
0
Fork 0
Code Issues 25 Pull Requests Actions Packages Projects Releases Wiki Activity
260 Commits 1 Branch 36 Tags
21e7d6c74b0ab3420241756629e3b25fda3d22e6
Commit Graph
4 Commits
Author SHA1 Message Date
mathiasandClaude Opus 4.8 9c2a04406b feat(llm): usage hook to surface token counts (ADR-030, #15) 
WithUsageHook callback fires with model + prompt/completion tokens parsed from the
response usage block. Keeps the copied stdlib-only llm package decoupled from
metrics (ADR-004) — the caller wires it to internal/metrics. TDD covered.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-12 08:36:02 +02:00
mathiasandClaude Opus 4.8 e9b5a3f3e7 feat(summarizer): resilient endpoint chain with local→cloud fallback (ADR-022)
CI / Lint / Test / Vet (push) Successful in 11s
Details
CI / Build & Import (push) Successful in 10s
Details 
The first friendly-pilot live run produced zero summaries: koala/phi4-mini hit
three silent failure modes — 8k context overflow on long transcripts (HTTP 400),
intermittent malformed JSON (highlights as a bare string), and no fallback wired
at all (summarizer.New(primary, nil)).

Keep phi4-mini as the fast primary and add resilience around it:

- Ordered endpoint chain (summarizer.NewChain): phi4-mini → koala/phi4-14b
  (local) → berget/mistral-small (worst-case external). All reached through the
  one LiteLLM gateway by alias.
- A parse failure now advances the chain like a transport error — the old
  Primary→Fallback shape returned the parse error without trying anyone else.
- Tolerant parse: highlights/takeaways coerce string→[]string, absorbing the
  common small-model quirk without spending a fallback round-trip.
- Transcript truncation (TAPIR_MAX_TRANSCRIPT_CHARS=18000) prevents the overflow
  rather than recovering from it; validated to fit phi4-mini's 8k window.
- Bounded completion budget (TAPIR_SUMMARY_MAX_TOKENS=1500) — the old 8192 budget
  itself contributed to the overflow.

Local-first guarantee preserved by ordering: external endpoint is tried only
after every local one fails. TAPIR_CLOUD_FALLBACK_MODEL="" disables it entirely
for client/NDA deployments.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-10 08:36:16 +02:00
mathiasandClaude Opus 4.8 c40b46b661 fix(llm): send generous max_tokens on every request 
The copied OpenAI-compatible client sent no max_tokens. Thinking models
(qwen3, deepseek-r1) spend their budget on the reasoning trace and return
EMPTY content when max_tokens is unset, which the summarizer treats as an
error. ADR-004 says change Tapir's copy rather than the hyperguild upstream,
so set a generous default (8192) leaving room for both reasoning and output.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-02 20:55:49 +02:00
mathiasandClaude Opus 4.8 0aeb3aa99e feat(adapters): copy stdlib llm package (Client + Router) per ADR-004 
Copy hyperguild/ingestion/internal/llm into internal/adapters/llm and own it.
Tapir owes that repo nothing at the dependency level — no module dep added.
Router gives the local-Primary -> BYO-Fallback path needed for ai_routing.feature.

Copied tests rewritten from testify to stdlib testing to keep go.mod
dependency-free (repo has zero deps; acceptance tests are stdlib too).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-02 17:01:56 +02:00