# Mubit

Category: Decision attribution and calibration for AI agents
Website: https://mubit.ai/

Attribute decisions and calibrate agents at scale
Mubit links each outcome to the agent decision that caused it, then uses those findings to change how agents handle similar tasks. Attribute decisions across your existing models and calibrate open-weight models with versioned, reversible adapters.

## What Mubit does
Mubit connects outcomes to the agent decisions that caused them, then uses those findings to change how agents handle similar tasks. Attribution works across models, including GPT and Claude. Adapter calibration is for open-weight models; updates are versioned and reversible.

## 01 / How it works
Plug in once. Get back better decisions. In the illustrative SRE scenario, the agent restarts a database without checking recent deploys. Errors return because the cause was a bad deploy. Mubit attributes the failed outcome to that decision. After calibration, the agent checks recent deploys first and rolls back the bad deploy.
Outcome attribution: https://docs.mubit.ai/patterns/outcome-attribution-loop

## 02 / Attribute
Connect each outcome to the decision that caused it, using the run's context and trace. Compare runs to identify the choices behind success and failure. Attribution works across models, including GPT and Claude.
Guide: https://docs.mubit.ai/patterns/outcome-attribution-loop

## 03 / Calibrate
Turn attributed findings into lessons, then into adapters that change how an open-weight model makes similar decisions. Calibration updates are versioned and reversible. This is distinct from attributing outcomes on hosted models.
Documentation: https://docs.mubit.ai/

## 04 / Mubit
Capture. Attribute. Calibrate.
1. Capture every decision, its context and its outcome.
2. Link each outcome to the decision that caused it, across runs.
3. Turn the finding into a lesson, then into a versioned, reversible model adapter.
Provider support and capture behavior differ across direct API calls, SDKs and framework helpers. Check the provider matrix before integrating.
Provider matrix: https://docs.mubit.ai/sdk/llm-providers
Framework adapters: https://docs.mubit.ai/sdk/framework-integrations

## 05 / Built for your agent stack
Call the API directly or use an SDK or framework helper. Start with one workflow in an isolated development scope, use environment-based credentials, capture the evidence IDs your agent used, and report a verified outcome with `record_outcome()`. Keep your existing production safeguards.
- Quickstart: https://docs.mubit.ai/getting-started
- Coding-agent setup: https://docs.mubit.ai/recipes/agent-memory-skill
- Run scopes: https://docs.mubit.ai/concepts/runs-and-scopes

## Examples and limits
- The homepage animations are conceptual illustrations, not recorded runs or measured benchmarks. The library artwork represents accumulated experience.
- A lesson must be relevant and authorized before reuse. Improvement is not guaranteed.
- Your application verifies outcomes. An agent saying it succeeded is not enough.
- Define behavior for empty retrieval, timeouts and stale lessons. Keep hard safety constraints in application code.
- Archival changes retrieval eligibility; deletion is a separate operation.
- Evaluate on held-out tasks with the same model, tools and budgets. Measure verified task success, repeated failures, latency and cost, including unsuccessful attempts.
- Confirm pricing, trial terms, regions, retention, deletion and deployment options with the team. No universal latency or measured performance uplift is claimed here.

## Get started or talk to the team
- Get started: https://console.mubit.ai/sign-in?redirect_url=/app/projects
- Documentation: https://docs.mubit.ai/
- Talk to someone: https://mubit.ai/talk-to-an-engineer/

## Further reference
- Retrieval semantics: https://docs.mubit.ai/concepts/retrieval-semantics
- Memory lifecycle: https://docs.mubit.ai/patterns/utility-weighted-forgetting
- Staleness: https://docs.mubit.ai/patterns/bi-temporal-memory
- Memory-backed handoffs: https://docs.mubit.ai/patterns/memory-backed-handoff
- Changelog: https://docs.mubit.ai/changelog