Skip to main content

Prompt caching

Prompt caching means the model provider reuses unchanged prompt prefixes — usually system/developer instructions and other stable context — across turns instead of re-processing them every time. The first matching request writes cache tokens (cacheRead/cacheWrite in usage surfaces); later matching requests read them back. This lowers token cost, speeds up responses, and makes long-running sessions more predictable. WednesdayAI controls caching through model params, context pruning, and the shape of the assembled system prompt. This page covers every knob that affects prompt reuse, plus the diagnostics that prove whether caching is working.

Primary knobs

cacheRetention

Set cache retention on model params (values: none | short | long):
Per-agent override:
Config merge order:
  1. agents.defaults.models["provider/model"].params
  2. agents.list[].params (matching agent id; overrides by key)
Legacy cacheControlTtl values are still accepted and mapped: 5m → short, 1h → long. Prefer cacheRetention for new config.

contextPruning.mode: "cache-ttl"

Prunes old tool-result context after cache TTL windows so post-idle requests do not re-cache oversized history:

Heartbeat keep-warm

Heartbeat keeps cache windows warm and reduces repeated cache writes after idle gaps:
Per-agent heartbeat is supported at agents.list[].heartbeat. See Heartbeat.

System prompt shape

Prompt caching works best when stable shared content appears before volatile content. WednesdayAI keeps identity, generated tool and skill indexes, stable workspace guidance, and stable Project Context before the cache boundary. Runtime details, channel-specific owner context, reply mechanics, and heartbeat state appear after it. The boundary is controlled by agents.defaults.systemPrompt.cache.boundary (overridable per agent at agents.list[].systemPrompt.cache.boundary):
  • auto (default) — insert the cache boundary between stable and volatile prompt sections.
  • off — render one continuous system prompt without the boundary marker.
When auto is active, the assembler inserts the HTML comment <!-- WEDNESDAYAI_CACHE_BOUNDARY --> between the stable prefix and volatile suffix, and closes Project Context blocks with <!-- WEDNESDAYAI_PROJECT_CONTEXT_END -->. The marker is a structural ordering aid — it groups stable content first so prefix-caching providers get longer reusable prefixes; it is not itself a provider cache-control breakpoint.
Renaming the marker text (an upgrade) changes every cached prompt prefix. After a build that renames these markers, expect a one-time cold cache — extra cacheWrite and latency on the first turn per session and model. No operator config change is required.
Use agents.defaults.systemPrompt.sections.projectContext.files for stable, ordered workspace context, and keep high-churn files out of the stable prefix. A cache-friendly shape is: stable identity and policy first, compact tool and skill listings next, stable project files next, then volatile runtime/session information. Use /context list in a chat session to confirm the effective preset, cache boundary, and Project Context file modes.

System prompt byte stability

Section ordering alone is not enough: prefix caching only pays off when the stable prefix is byte-identical from turn to turn. WednesdayAI keeps the system prompt byte-stable across trigger kinds within a session — a heartbeat or cron turn that shares a session with user messages renders the same system prompt bytes as a user turn. Heartbeat-specific instructions (such as HEARTBEAT.md content in a shared session) travel in the trigger message rather than swapping the system prompt, so scheduled turns do not churn the cached prefix. Two things can still legitimately change the system prompt within a session:
  • A real configuration or workspace change (edited bootstrap files, model or tool changes, new skills).
  • An extension or plugin that injects per-turn dynamic content into the system prompt. Prefer delivering per-turn data as user-role context instead; system-prompt injection should be session-stable.
Every turn computes a hash of the assembled system prompt and compares it with the previous turn in the same session. A change increments the flip counters below and logs a rate-limited warning (system prompt hash flipped within session, at most once per 10 minutes per session) with the before and after hashes and sizes.

Provider behavior

Tuning patterns

Mixed traffic (recommended default) — keep a long-lived baseline on your main agent, disable caching on bursty notifier agents:
Cost-first baseline — set cacheRetention: "short" as the baseline, enable contextPruning.mode: "cache-ttl", and keep heartbeat below your TTL only for agents that benefit from warm caches.

Cache diagnostics

WednesdayAI exposes dedicated cache-trace diagnostics for embedded agent runs.

diagnostics.cacheTrace

Defaults: filePath is <state dir>/logs/cache-trace.jsonl (state dir is ~/.openclaw unless OPENCLAW_STATE_DIR is set); includeMessages, includePrompt, and includeSystem default to true. Env toggles for one-off debugging: Cache trace events are JSONL and include staged snapshots such as session:loaded, prompt:before, stream:context, and session:after. Per-turn cache token impact is visible in normal usage surfaces via cacheRead and cacheWrite — for example the /usage full chat command and session usage summaries.

diagnostics.modelPayloadLog

For provider-level request/usage evidence, the model payload log records JSONL events at the real stream boundary:
Defaults: enabled: false; usageSummaries: true; rawPayloads: false; file <state dir>/logs/model-payload.jsonl. Env equivalents: OPENCLAW_MODEL_PAYLOAD_LOG, OPENCLAW_MODEL_PAYLOAD_LOG_FILE (the older OPENCLAW_ANTHROPIC_PAYLOAD_LOG[_FILE] names still work but log a deprecation warning).
rawPayloads: true writes provider request bodies to disk. Uncaptured fields are replaced with [stripped] based on the same captureContent policy used by OTel export (default: messages, system, and tool definitions stripped unless explicitly enabled), and output is redacted and size-bounded — but treat the resulting file as sensitive and delete it when the investigation ends.

System prompt stability metrics

When the diagnostics-otel plugin is enabled, two metrics track whether system prompts stay byte-stable within a session (see System prompt byte stability): Interpretation:
  • Occasional flips after a config, model, or workspace-file change are expected — the first turn after the change pays a cold cache, then stability resumes.
  • A flip on every heartbeat or cron turn means a scheduled trigger renders a different system prompt than user turns. Core heartbeat and cron paths are byte-stable; the usual cause is a plugin or extension injecting per-turn dynamic content into the system prompt. Move that content to user-role context.
  • A flip on every turn points at volatile content in the stable prefix (timestamps, random ordering, per-turn scores). Check the prevChars/nextChars sizes in the rate-limited gateway warning to narrow down which section changed.
The same comparison emits a session.system-prompt.flip diagnostic event, so custom diagnostic consumers can alert on flips without scraping metrics. Metric setup is covered in OTel and Langfuse diagnostics.

Quick troubleshooting