Skip to main content
Coming soon. Autoresearch is in research preview and not yet available. Requires a Simmer Pro plan.
Autoresearch lets your agent optimize its own trading skills. It runs experiments — changing config values, measuring results over real trading cycles, and keeping changes that improve performance. Think of it as automated A/B testing for your trading strategy.

​
How it works

init_experiment → run_experiment (N cycles) → log_experiment → repeat
  1. Init — Pick a skill and a hypothesis (e.g., “lower min_edge from 0.05 to 0.03”)
  2. Run — Execute the skill with the new config for several trading cycles
  3. Log — Record results (P&L, win rate, trade count) and decide: keep or revert
  4. Repeat — Try the next hypothesis
Your agent drives the loop — autoresearch provides the tools, your agent provides the reasoning.

​
Install

openclaw plugins install simmer-autoresearch

​
Config

OptionDefaultDescription
apiKey$SIMMER_API_KEYSimmer API key. Optional if env var is set.
apiUrlhttps://api.simmer.marketsAPI base URL
maxExperiments50Max experiments per session. Prevents runaway loops. 0 = unlimited.
Example plugins.json: 
{
  "simmer-autoresearch": {
    "maxExperiments": 30
  }
}

​
Tools

The plugin registers three tools your agent can call:

​
init_experiment

Start a new experiment session for a skill.
ParameterRequiredDescription
skill_slugYesClawHub slug of the skill to optimize (e.g., polymarket-fast-loop)
hypothesisYesWhat you’re testing (e.g., “increase volatility threshold to 0.7”)
config_changesNoKey-value pairs of config changes to apply

​
run_experiment

Execute the skill under test. Call this repeatedly (once per trading cycle) to collect data. Returns the skill’s output from that cycle — trades made, signals found, errors hit.

​
log_experiment

Record experiment results and decide whether to keep or revert.
ParameterRequiredDescription
statusYeskeep, discard, or crash
metricsNoObserved metrics (P&L, win rate, trade count, etc.)
notesNoAgent’s reasoning for the decision

​
Commands

CommandDescription
/autoresearch statusCurrent skill, experiment count, keep rate, budget remaining, pause state
/autoresearch resetClear state and start fresh (clears pause if paused)

​
Safety features

​
Crash protection

  • Baseline crash — If the very first experiment in a session crashes, autoresearch pauses automatically. This usually means the skill is misconfigured.
  • Consecutive crashes — 3 crashes in a row triggers auto-pause. Your agent can’t run more experiments until the issue is investigated.
  • Recovery — init_experiment (new session) or /autoresearch reset clears the pause.

​
Budget caps

Experiments are capped at maxExperiments (default 50) per session. At 80% of the cap, your agent gets a warning. At the limit, run_experiment is blocked. Set maxExperiments: 0 to disable the cap (not recommended for unattended agents).

​
Experiment persistence

Results are saved in two places:
  • Local JSONL — {stateDir}/autoresearch/experiments.jsonl for offline access
  • Dashboard API — Synced to your Simmer dashboard for viewing experiment history
Git auto-commits on keep decisions so you can track what changed and roll back if needed.

​
Example workflow

A typical autoresearch session looks like: 
Agent: "I'll optimize polymarket-fast-loop. Hypothesis: lowering min_edge
       from 0.05 to 0.03 will catch more opportunities."

→ init_experiment(skill_slug="polymarket-fast-loop",
                  hypothesis="lower min_edge to 0.03",
                  config_changes={"min_edge": 0.03})

→ run_experiment() × 5 cycles

Agent: "5 cycles complete. 3 trades found (vs 0 with old config).
       Net P&L: +$2.40. Keeping this change."

→ log_experiment(status="keep",
                 metrics={"pnl": 2.4, "trades": 3, "cycles": 5},
                 notes="3x more opportunities with lower edge threshold")

​
API endpoints

These endpoints power the plugin’s server-side sync. You don’t call them directly — the plugin handles it.
EndpointDescription
GET /api/sdk/autoresearch/experimentsList experiment history
POST /api/sdk/autoresearch/experimentsSync experiment results
GET /api/sdk/autoresearch/experiments/latestLatest experiment