Add smart-client retry/queue integration for Agent Zero #71

New issue

Open

opened 2026-09-12 19:54:49 +00:00 by nsaspy · 0 comments

nsaspy commented

2026-09-12 19:54:49 +00:00

Owner

Goal

Let Agent Zero cooperate with llm-log backpressure instead of treating temporary saturation as a model failure.

Scope

Add an llm-log/provider plugin path for Agent Zero rather than patching core scheduling logic into Agent Zero.
Honor 429 Too Many Requests and Retry-After.
Retry with bounded jitter/backoff and an overall request deadline.
Surface queued/retrying state distinctly from provider/model errors.
Allow a configurable client-side queue for workloads where waiting is acceptable.
Keep llm-log authoritative for provider-wide concurrency so multiple Agent Zero workers cannot multiply the effective provider limit.

Tests

429 then success.
repeated 429 until deadline.
malformed/missing Retry-After fallback.
cancellation while waiting.
no duplicate execution after retry.

Implement after #70 establishes the server-side contract.

## Goal Let Agent Zero cooperate with llm-log backpressure instead of treating temporary saturation as a model failure. ## Scope - Add an llm-log/provider plugin path for Agent Zero rather than patching core scheduling logic into Agent Zero. - Honor `429 Too Many Requests` and `Retry-After`. - Retry with bounded jitter/backoff and an overall request deadline. - Surface queued/retrying state distinctly from provider/model errors. - Allow a configurable client-side queue for workloads where waiting is acceptable. - Keep llm-log authoritative for provider-wide concurrency so multiple Agent Zero workers cannot multiply the effective provider limit. ## Tests - 429 then success. - repeated 429 until deadline. - malformed/missing Retry-After fallback. - cancellation while waiting. - no duplicate execution after retry. Implement after #70 establishes the server-side contract.

nsaspy referenced this issue

2026-09-12 23:06:24 +00:00

Rate admission: enforce time-based quotas shared across client/provider aliases #75

nsaspy referenced this issue from a pull request that will close it,

2026-09-12 23:21:00 +00:00

Rate-budget and outbound-profile foundation (stacked on #74) #76

nsaspy referenced this issue

2026-09-12 23:21:00 +00:00

Rate admission: time-based quotas shared across client aliases #77

nsaspy referenced this issue

2026-09-21 20:12:52 +00:00

Add provider-wide request scheduling and 429 backpressure #74