feat: replace placeholder token estimates with real counts from agentic-kit

[pyramation]

Summary

Replaces all placeholder token estimates with real provider counts from @agentic-kit/[email protected].

Chat tokens (from earlier commits): OllamaAdapter.stream() returns Usage with real prompt_tokens, completion_tokens, reasoning_tokens, cache_read_tokens, cache_write_tokens. ChatFunction now returns ChatResult { content, usage } instead of string.

Embedding tokens (new commit): generateEmbedding() now returns EmbeddingResult { embedding, promptTokens } from Ollama's /api/embed endpoint (prompt_eval_count). Removes the ~4 chars/token placeholder estimate in metering.ts.

Changes

• Bump @agentic-kit/ollama from ^1.2.1 to ^2.0.0 (breaking: generateEmbedding returns EmbeddingResult)
• Add EmbeddingResult type to graphile-llm/types.ts, update EmbedderFunction return type
• metering.ts: replace Math.ceil(text.length / 4) with real promptTokens from provider
• metering.ts: record prompt_tokens in usage metadata and rawUsage
• All embedding consumers updated to destructure { embedding, promptTokens }
• CLI codegen template: extracts .embedding internally (keeps Promise<number[]> API for generated code)
• Tests updated for new EmbeddingResult shape

Review & Testing Checklist for Human

• Verify EmbeddingResult type flows correctly through metering → billing pipeline (check record_usage receives real promptTokens not estimated values)
• Test with a real Ollama instance: buildEmbedder({ provider: 'ollama' }) should return { embedding: number[], promptTokens: number } with promptTokens > 0
• Verify CLI --auto-embed still works (codegen template returns number[] not EmbeddingResult)
• Check that quota-exceeded path in metering.ts logs inputTokens: 0 (no estimate since embedding wasn't called)

Notes

• Depends on @agentic-kit/[email protected] (published, PR Feat/plan #11 merged in agentic-kit)
• The as unknown as EmbedderFunction cast in metering-plugin.ts is intentional — the metered wrapper returns number[] | null (stripping token count after recording it), while llmEmbedder on Build is typed as EmbedderFunction. Downstream plugins (text-search, text-mutation) already cast to (text: string) => Promise<number[] | null>.
• Some files include trailing-comma lint fixes from a previous session (no functional changes)

Link to Devin session: https://app.devin.ai/sessions/2b5a29d83d3f478e8d3d972653b4879c
Requested by: @pyramation

[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 and CI monitoring

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	@​pgpm/​metaschema-schema@​0.18.0
	@​pgpm/​metaschema-modules@​0.21.1
	@​pgpm/​database-jobs@​0.18.0
	@​pgpm/​inflection@​0.18.0
	@​pgpm/​types@​0.18.0
	@​pgpm/​verify@​0.18.0
	@​launchql/​styled-email@​0.1.0
	@​pgpm/​services@​0.18.0
	@​jest/​test-sequencer@​30.4.1
	@​launchql/​mjml@​0.1.1
	@​types/​jest-in-case@​1.0.9
	@​constructive-io/​fetch@​1.1.1
	@​types/​express@​5.0.6
	@​types/​pg-copy-streams@​1.2.5
	@​types/​supertest@​7.2.0
	@​inquirerer/​utils@​3.3.7
	@​types/​multer@​2.1.0
	@​types/​pg@​8.20.0
	@​types/​smtp-server@​3.5.13
	@​agentic-kit/​ollama@​2.0.0
	@​types/​react-dom@​19.2.3
	@​types/​shelljs@​0.10.0
	@​inquirerer/​test@​1.4.1
	@​types/​nodemailer@​7.0.11
	@​dataplan/​json@​1.0.0
	@​types/​jest@​30.0.0
	@​types/​react@​19.2.15
	@​babel/​generator@​7.29.1
	@​constructive-io/​noble-hashes@​0.2.1
	@​types/​request-ip@​0.0.41
	@​types/​graphql-upload@​8.0.12
	@​types/​cors@​2.8.19
See 14 more rows in the dashboard

View full report