Reflect turns AI coding sessions into shared evidence for better procedures, smarter capacity decisions, and reusable practices that scale across agents, repositories, and teams.
From Agent Work to Measured Improvement
Reflect connects sessions, workspaces, repositories, tools, skills, outcomes, and local memory so evidence can explain what happened, guide better procedures, and show what is ready to scale.
Follow prompts, responses, tool and MCP calls, failures, retries, subagents, token use, cost, and outcomes on one synchronized timeline—even when evidence arrives through native sessions and hooks.
Move beyond isolated transcripts. Stable hook, native-trace, workspace, repository, MCP, and delegation identities connect sessions to the work they actually performed.
Detect repeated inputs, failed recovery, broad exploration, late verification, and patterns that consume time without changing state or improving the work.
Review evidence-backed workflow and skill proposals, inspect the exact change, install approved versions into the right repository or agent, and retain their history.
Compare later sessions against the original cohort before claiming that a workflow, skill, or rule improved the outcome.
Ask Through Your Agent
Start with a normal engineering or business question. The agent uses Reflect's local evidence to explain the work, recommend follow-up actions, and prepare evidence for personal learning, team decisions, or internal requests.
Let the agent connect exact usage to past work, separate productive growth from avoidable waste, recommend improvements, and prepare a reviewed internal request.
$reflect-usageExplain our AI budget need and what should improve.
Open the synchronized conversation, timeline, tool activity, subagents, workspace context, failures, and cost evidence.
$reflectExplain why this session stalled or failed.
Rank repeated problems by impact, then inspect the exact observations and source sessions behind each recommendation.
$reflectRank what we should improve first.
Review bounded loop evidence and let an agent author one pending workflow packaged as a skill for deliberate approval.
$reflect-skillsFind repeated behavior worth turning into a reusable skill.
Retrieve reviewed workflows, linked evidence, and folder-scoped memory instead of relying on generic advice.
$reflectExplain how an agent should work in this repository.
Search local, provenance-aware memory for a repository decision, release practice, debugging path, or prior constraint.
$reflectFind what we already learned about the release gate.
Check hook installation, native telemetry, local files, gateway state, detected agents, pricing, and integration support.
$reflectCheck whether agent telemetry and integrations are healthy.
Personal Insight to Organizational Learning
Start with a personal question about one session, then follow the same local evidence into shared procedures, capacity decisions, and practices that scale across the organization.
Personal reflection
Organizational reflection
How It Works
Connect once, ask through your agent, and use the visual UI whenever you want the complete evidence and review path.
Reflect configures local OTLP collection and reads supported native session stores without forcing every agent through one vendor gateway.
reflect setup
Ask natural questions about usage, failures, repeated work, repository guidance, and improvement. Reflect gives the agent bounded, provenance-aware evidence.
Use Reflect to explain this work.
Open the UI to inspect sessions and costs, review exact workflow or skill changes, approve or reject them, and see their later impact.
reflect
Local-First Architecture
Reflect keeps its evidence graph and workflow ledger in local SQLite. Its read-only MCP gives agents scoped context with provenance, while optional memory providers such as OMEGA remain independently installed and clearly separated from Reflect-verified evidence.
Memory, With Boundaries
Reflect keeps evidence and provenance in its own local ledger, then connects to optional memory systems without confusing external recall with verified session facts.
Start Locally
Connect Reflect once, ask through your coding agent, and open the visual dashboard whenever you want the complete evidence and review path.
$ pipx install o11y-reflect
$ reflect setup
# Then ask in your coding agent
› Use Reflect to explain why our team needs more AI budget.
# Open the visual overview
$ reflectAsk. Review. Improve.
Get the immediate explanation in the conversation, then use the visual UI to inspect the source evidence, control what becomes reusable, and measure what helped.