Files
0pt/tests/document-workbench/decision-ollama-client.test.ts
admin 8b571cf2f4
CI (SIAX Cloud) / integration (push) Skipped
CI (SIAX Cloud) / sonar (push) Skipped
CI (SIAX Cloud) / sonar (pull_request) Skipped
masterplan-lock / check (pull_request) Successful in 5s
Security Scan (SIAX Cloud) / security (push) Successful in 14s
CI (SIAX Cloud) / security (push) Successful in 17s
masterplan-lock / check (push) Successful in 17s
CI (SIAX Cloud) / security (pull_request) Successful in 20s
Security Scan (SIAX Cloud) / security (pull_request) Successful in 21s
CI (SIAX Cloud) / integration (pull_request) Successful in 54s
CI (SIAX Cloud) / quality (push) Successful in 1m24s
CI (SIAX Cloud) / quality (pull_request) Successful in 1m29s
fix(decision-ollama): use Ollama's native JSON mode without thinking
The 9B model spent its whole 4000-token budget on reasoning and came back
model_incomplete. Call /api/chat with format=json, think=false and num_predict,
and read prompt_eval_count/eval_count/done_reason. Structured extraction no
longer needs chain-of-thought.
2026-11-02 17:51:09 +01:00

124 lines
4.0 KiB
TypeScript

import { describe, expect, it } from 'vitest';
import {
DecisionOllamaGatewayClient,
assertTrustedDecisionOllamaUrl,
} from '../../src/document-workbench/decision-ollama-client';
import { Inf0OutcomeUnknownError } from '../../src/document-workbench/inf0-gateway-client';
import { createInf0Gateway } from '../../src/document-workbench/inf0-gateway';
const context = {
tenantId: 't_0pt',
projectId: 'prj_default',
actorId: 'user_1',
capability: '0pt.document.structure.v1',
correlationId: 'corr-1',
};
const admission = {
tenantId: 't_0pt',
projectId: 'prj_default',
jobId: 'job-1',
modelId: 'hermes-qwen3.5:9b',
dataClass: 'low',
maxInputTokens: 1000,
maxOutputTokens: 200,
maximumCredits: 0,
};
const policy = {
tenantId: 't_0pt',
projectId: 'prj_default',
allowedModelIds: ['hermes-qwen3.5:9b'],
allowedDataClasses: ['low'],
destination: 'internal' as const,
externalTransferAllowed: false,
maxInputTokens: 1000,
maxOutputTokens: 200,
};
const messages = [
{ role: 'system' as const, content: 'system' },
{ role: 'user' as const, content: 'data' },
];
function client(fetchImpl: typeof fetch, model = 'hermes-qwen3.5:9b') {
return new DecisionOllamaGatewayClient({
baseUrl: 'http://100.96.133.104:11434',
model,
fetchImpl,
});
}
describe('decision server Ollama gateway', () => {
it('accepts tailnet/RFC1918 HTTP but refuses a public HTTP endpoint', () => {
expect(() => assertTrustedDecisionOllamaUrl('http://100.96.133.104:11434')).not.toThrow();
expect(() => assertTrustedDecisionOllamaUrl('http://192.168.1.10:11434')).not.toThrow();
expect(() => assertTrustedDecisionOllamaUrl('https://ai.example.test')).not.toThrow();
expect(() => assertTrustedDecisionOllamaUrl('http://8.8.8.8:11434')).toThrow();
expect(() => assertTrustedDecisionOllamaUrl('http://user:pass@100.96.133.104:11434')).toThrow();
});
it('maps a native Ollama chat response into the INF0 receipt', async () => {
const c = client(async (url, init) => {
expect(String(url)).toMatch(/\/api\/chat$/);
const body = JSON.parse(String(init?.body));
expect(body.format).toBe('json');
expect(body.think).toBe(false);
return Response.json({
model: 'hermes-qwen3.5:9b',
message: { role: 'assistant', content: '{"value":{}}' },
done: true,
done_reason: 'stop',
prompt_eval_count: 12,
eval_count: 8,
});
});
await expect(
c.complete({ context, admission, policy, reservation: undefined, inputTokens: 12, messages }),
).resolves.toMatchObject({
status: 'model_stopped',
modelExecuted: 'hermes-qwen3.5:9b',
providerKind: 'local',
totalTokens: 20,
});
});
it('refuses an external destination and malformed usage', async () => {
const c = client(async () =>
Response.json({
model: 'hermes-qwen3.5:9b',
message: { role: 'assistant', content: 'x' },
done: true,
done_reason: 'stop',
prompt_eval_count: 1,
eval_count: 5000,
}),
);
await expect(
c.complete({
context,
admission,
policy: { ...policy, destination: 'external' },
reservation: undefined,
inputTokens: 1,
messages,
}),
).rejects.toThrow(/denied/i);
await expect(
c.complete({ context, admission, policy, reservation: undefined, inputTokens: 1, messages }),
).rejects.toBeInstanceOf(Inf0OutcomeUnknownError);
});
it('is preferred over the INF0/LiteLLM path when configured', () => {
expect(
createInf0Gateway({
DECISION_OLLAMA_URL: 'http://100.96.133.104:11434',
DECISION_OLLAMA_MODEL: 'hermes-qwen3.5:9b',
INF0_GATEWAY_LITELLM_KEY: 'sk-test',
} as unknown as NodeJS.ProcessEnv),
).toBeInstanceOf(DecisionOllamaGatewayClient);
expect(
createInf0Gateway({
INF0_GATEWAY_LITELLM_KEY: 'sk-test',
} as unknown as NodeJS.ProcessEnv),
).not.toBeInstanceOf(DecisionOllamaGatewayClient);
});
});