# Memrail documentation

> Memrail provides deterministic steering and control for AI agents and automated systems. AMI (Adaptive Memory Intelligence) is the decision engine. These guides use the v2 trigger DSL; HTTP routes retain the /v1 prefix.

## Read first: integration contract
- [Integration invariants](https://docs.memrail.com/concepts/index.md#integration-invariants): Named decision_point bindings, evaluation, execution, lifecycle, outcomes, and idempotency.
- [EMU JSON Schema](https://docs.memrail.com/schemas/emu.schema.json): JSON Schema 2020-12 for EMU registration structure. Validate each JSONL object separately, then run strict preflight for semantic checks.
- [Diagnostic codes](https://docs.memrail.com/reference/diagnostic-codes.json): Validation, suppression, and core HTTP error identifiers with suggested responses.
- [Runnable quickstart](https://docs.memrail.com/examples/quickstart.py): Complete Python program using the SDK executor; setup and promotion instructions in the getting-started guide.
- [Quickstart policy](https://docs.memrail.com/examples/emus.jsonl): Matching named EMU, ready for JSONL plan/apply.

## Guides
- [Deterministic steering and control for AI agents](https://docs.memrail.com/index.md): Memrail gives AI agents and automated systems explicit, testable control at every consequential action—without putting probabilistic inference in charge of policy.
- [Install the Memrail skill](https://docs.memrail.com/agents/index.md): Give Codex, Claude Code, or OpenCode Memrail expertise. Copy a setup request for your agent, or install the complete skill with one terminal command.
- [Why declarative agent control matters: one reviewable refund change](https://docs.memrail.com/declarative-agent-control/index.md): Make agent-written changes reviewable with explicit Memrail policies. Follow a refund example through bounded edits, tests, and application execution checks.
- [Start from scratch](https://docs.memrail.com/getting-started/index.md): Run a complete Memrail example: install the SDK, bind a policy to a named decision point, validate in shadow, promote, execute a local tool, and acknowledge success.
- [Add Memrail to an existing project](https://docs.memrail.com/existing-project/index.md): Introduce deterministic agent control incrementally by discovering hidden authority, selecting one consequential seam, preserving current behavior, and migrating policy in shadow.
- [Core concepts](https://docs.memrail.com/concepts/index.md): Understand Memrail's runtime model: decision points, ATOMs, EMUs, actions, policies, project coherence, reachability, and traces.
- [AI agent steering and control patterns](https://docs.memrail.com/agent-control/index.md): Place deterministic Memrail control points around model reasoning, tool calls, human escalation, and material outcomes without replacing the agent framework.
- [Python SDK](https://docs.memrail.com/python/index.md): Configure AsyncAMIClient, build state and tag context, invoke named decision points, ingest events, trace decisions, and handle Memrail actions in Python.
- [TypeScript SDK](https://docs.memrail.com/typescript/index.md): Configure AMIClient, build typed context, invoke named decision points, emit events, trace policy selection, and dispatch Memrail actions in TypeScript.
- [Trigger DSL](https://docs.memrail.com/triggers/index.md): Write deterministic Memrail conditions over state, bounded tags, and event history with reachability, scoping, and retention in mind.
- [Write and validate EMUs](https://docs.memrail.com/emus/index.md): Design production-ready Executable Memory Units with reachable triggers, connected actions, safe policy controls, JSONL source, and server validation.
- [Audit an application's decision topology](https://docs.memrail.com/topology-audit/index.md): A read-only audit method for finding hidden AI agent authority, mapping decision points and atom contracts, proving trigger reachability, and ranking control gaps.
- [Recursive self-improvement with Hindsight](https://docs.memrail.com/hindsight/index.md): Improve agent steering and control through a reviewed feedback loop. Use Memrail Hindsight to analyze outcomes, propose policy changes, and evaluate the next version.
- [Production workflow](https://docs.memrail.com/production/index.md): Manage Memrail policy as reviewed JSONL, verify deployed scope and lifecycle, evaluate without live interference, and control application rollout and rollback.
- [Troubleshooting](https://docs.memrail.com/troubleshooting/index.md): Diagnose Memrail policies that do not fire, actions that cannot execute, validation failures, event mismatches, configuration errors, and unsafe fallbacks.

## Complete corpus
- [All documentation](https://docs.memrail.com/llms-full.txt): Every public documentation page in one Markdown document.
- [Install the Memrail skill](https://docs.memrail.com/agents/install-skill.md): Instructions an agent can follow to install the skill, with optional SDK setup.