# Memtro

> Memory infrastructure & AI Orchestration for your company. An OpenAI-compatible API in front of the tools a team already uses (750 connectors, reading by default and acting where you grant it: Gmail, Outlook, Google and Outlook calendars, Slack, LinkedIn (personal), Microsoft Teams, Granola, Drive, SharePoint, Notion, Confluence, HubSpot, Salesforce, Pipedrive, Twenty, Help Scout, Zendesk, Intercom, Freshdesk, Xero, QuickBooks, Stripe, Felloh, Starling Bank, Shopify, Linear, Jira, Asana, monday.com, Airtable, GitHub, Sentry, CloudWatch, Postgres, MySQL, SQL Server, BigQuery, Snowflake, Google Analytics, Plausible, Google Search Console, BambooHR, AWS Athena, Zoho CRM, Dynamics 365, Mailchimp, Klaviyo, GitLab, PagerDuty, Datadog, Zoom, Google Workspace and Entra ID directories) and the models it already pays for (Anthropic, OpenAI, Google, xAI, OpenRouter, DeepSeek, OpenAI-compatible endpoints, Claude Code), with a shared knowledge-graph memory, connector tools that read unless an action is granted, model routing by difficulty, a multi-step agent mode, jobs that run on a schedule or start when one of 383 connected systems sends an event, and an MCP server.

Humans sign in at https://platform.memtro.com. Agents use an API key (`sk-memtro-…`) against https://platform.memtro.com/v1 or the MCP server at https://platform.memtro.com/api/mcp. See https://www.memtro.com/agents.md for a connection recipe.

## Docs

- [Getting started](https://www.memtro.com/docs/getting-started/): Sign in, add a model key, create an API key and send your first request, then connect a first system so answers come with citations.
- [Quickstart](https://www.memtro.com/docs/quickstart/): Your first request from curl, Python, JavaScript, OpenCode or Claude Code.
- [Agent mode](https://www.memtro.com/docs/agent-mode/): Multi-step investigations, cross-referencing, and how the loop is confined.
- [Billing, seats and leaving](https://www.memtro.com/docs/billing/): What the subscription covers, the tiers, what happens when somebody leaves, and cancelling versus deleting the organisation.
- [Code mode](https://www.memtro.com/docs/code-mode/): How agent mode writes a short script to read, filter and join across systems in one step, why a script can only read, and where it runs.
- [Commitments](https://www.memtro.com/docs/commitments/): Who promised what in a meeting, by when, and whether the other systems show it landed.
- [Connectors](https://www.memtro.com/docs/connectors/): Every connected system, what the model can do with it, and how to connect it.
- [Contradictions](https://www.memtro.com/docs/contradictions/): The things your business believes twice, differently — found by reading its own memory against itself, and corrected by answering them.
- [Cost control](https://www.memtro.com/docs/cost-control/): A token allowance a month for each model choice: when one runs out the cheaper one answers, and a level with nothing left refuses instead of costing more.
- [Decisions](https://www.memtro.com/docs/decisions/): The reasons behind the code, read from pull requests and threads, kept against the code and posted back when you ask.
- [Events and webhooks](https://www.memtro.com/docs/events/): A connected app tells Memtro something happened and a job starts: turning events on, the address and its secret, how deliveries are checked, the limits, and what a run started by an event may do.
- [Jargon](https://www.memtro.com/docs/jargon/): Your own vocabulary, as part of memory: the words your business uses in its own way and what each one points at, so every answer starts from your usage rather than the dictionary's.
- [Jobs](https://www.memtro.com/docs/jobs/): Work that runs on a schedule, once, or when a connected app sends an event: digests, monitors and reminders that run with their owner's tools and report back in a chat.
- [MCP server](https://www.memtro.com/docs/mcp/): Use Memtro's tools from any MCP client: Claude Code, Cursor, Claude Desktop and your own agents get memory and connectors over one endpoint.
- [Measures](https://www.memtro.com/docs/measures/): What memory knows about your numbers: the figures your business has agreed the meaning of, and the rule behind each, so a figure in an answer is yours rather than the obvious reading of the words.
- [Memory](https://www.memtro.com/docs/memory/): How Memtro learns, personal versus shared knowledge, history as of any date, and importing from ChatGPT and Claude.
- [Model routing](https://www.memtro.com/docs/routing/): The auto model: per-level ladders, the difficulty classifier, fallbacks, the coding ladder and the agent-mode floor, set per organisation, team or person.
- [Privacy and erasure](https://www.memtro.com/docs/privacy/): Answer a request to be forgotten: find everything held about a person, erase it across every store, and keep the receipt.
- [Procedures](https://www.memtro.com/docs/procedures/): The other half of memory: not what is true, but how the work is done here — so the assistant follows your sequence rather than inventing a sensible-looking one.
- [Prompt injection](https://www.memtro.com/docs/prompt-injection/): What happens when a ticket tells your assistant what to do: the rules Memtro applies outside the model, what they cost, and what they do not cover.
- [Single sign-on and provisioning](https://www.memtro.com/docs/identity/): Sign in through Microsoft Entra ID, Okta, OpenID Connect or SAML 2.0, verify your domains, and let SCIM manage joiners, leavers and groups.
- [Teams and permissions](https://www.memtro.com/docs/teams/): Admins and users, teams, who can use which shared connection, and what users may set up.
- [Time travel](https://www.memtro.com/docs/time-travel/): Read memory as of any date, follow how a fact changed, and ask what the company believed in July.
- [Usage statistics](https://www.memtro.com/docs/usage/): Tokens and requests by user, team, model and day, with CSV export, so the cost of every conversation is visible to the people who pay for it.
- [Tools reference](https://www.memtro.com/docs/tools/): Every memtro_ tool the model and MCP clients can call, with its inputs.
- [Troubleshooting](https://www.memtro.com/docs/troubleshooting/): Every error code the API returns, what it means and how to fix it, from a missing key to a level whose allowance ran out this month.

## API reference

- [API reference](https://www.memtro.com/docs/api/): Authentication, model ids, the memtro object, streaming, errors and limits, and an index of every endpoint: everything an OpenAI- or Anthropic-compatible client needs to talk to Memtro.
- [POST /v1/chat/completions](https://www.memtro.com/docs/api/chat-completions/): Answer a conversation in the OpenAI chat-completions shape, through memory, routing, connector tools and agent mode, streamed or in one reply.
- [POST /v1/responses](https://www.memtro.com/docs/api/responses/): The same pipeline in the OpenAI Responses shape, for clients such as the Codex CLI that speak only that wire.
- [POST /v1/messages](https://www.memtro.com/docs/api/messages/): The same pipeline in the Anthropic Messages shape, for Claude Code and the Anthropic SDKs.
- [POST /v1/messages/count_tokens](https://www.memtro.com/docs/api/count-tokens/): Count the input tokens of a Messages request: Anthropic's own count for an Anthropic model, otherwise an estimate marked as one.
- [POST /v1/embeddings](https://www.memtro.com/docs/api/embeddings/): Embed text through your own OpenAI or Google key, in the OpenAI embeddings shape.
- [GET /v1/models](https://www.memtro.com/docs/api/models/): The model ids this key can ask for, with context and output limits where they are known, in both OpenAI and Anthropic fields.
- [GET /v1/models/{model}](https://www.memtro.com/docs/api/models-retrieve/): One model by id, including ids that contain slashes.
- [GET /v1/memory](https://www.memtro.com/docs/api/memory-list/): Page through the memories this key can read, newest first, or search them by meaning or by words.
- [GET /v1/memory/{id}](https://www.memtro.com/docs/api/memory-retrieve/): Read one memory by id.
- [POST /v1/memory](https://www.memtro.com/docs/api/memory-import/): Hand Memtro a conversation it was not part of, and it learns from it as though it had been: no answer is generated.
- [DELETE /v1/memory/{id}](https://www.memtro.com/docs/api/memory-delete/): Delete one memory permanently.
- [GET /v1/entities/{id}](https://www.memtro.com/docs/api/entities-retrieve/): One entity from the knowledge graph, with its recorded values, its relationships and its memories.
- [GET /v1/usage](https://www.memtro.com/docs/api/usage/): What this key's holder has spent over a window, and the state of each capped allowance this month.
- [POST /api/mcp](https://www.memtro.com/docs/api/mcp/): Memtro as a stateless Streamable HTTP MCP server: memory and connector tools for any MCP client.
- [OpenAPI spec](https://www.memtro.com/openapi.yaml) (also as [JSON](https://www.memtro.com/openapi.json)), generated from the same reference

## Guides

- [Importing ChatGPT and Claude memory](https://www.memtro.com/docs/guide-import/), [Scheduling a job from chat](https://www.memtro.com/docs/guide-jobs/)

## Integrations

- [Claude Code](https://www.memtro.com/docs/guide-claude-code/), [Codex](https://www.memtro.com/docs/guide-codex/), [Cursor](https://www.memtro.com/docs/guide-cursor/), [Help Scout](https://www.memtro.com/docs/guide-help-scout/), [Memtro CLI](https://www.memtro.com/docs/guide-cli/), [Open WebUI](https://www.memtro.com/docs/guide-open-webui/), [OpenAI SDKs](https://www.memtro.com/docs/guide-openai-sdk/), [OpenCode](https://www.memtro.com/docs/guide-opencode/), [Pi](https://www.memtro.com/docs/guide-pi/), [Postgres and warehouses](https://www.memtro.com/docs/guide-postgres/), [Slack](https://www.memtro.com/docs/guide-slack/)

## Providers

- [Anthropic (Claude)](https://www.memtro.com/docs/guide-anthropic/), [DeepSeek](https://www.memtro.com/docs/guide-deepseek/), [Google Gemini](https://www.memtro.com/docs/guide-google/), [Meta (Muse Spark)](https://www.memtro.com/docs/guide-meta/), [OpenAI](https://www.memtro.com/docs/guide-openai/), [OpenAI-compatible endpoints](https://www.memtro.com/docs/guide-openai-compatible/), [OpenRouter](https://www.memtro.com/docs/guide-openrouter/), [xAI (Grok)](https://www.memtro.com/docs/guide-xai/)

## Other

- [Connectors](https://www.memtro.com/connectors/) (one landing page per connector, searchable), [Integrations](https://www.memtro.com/integrations/), [Use cases](https://www.memtro.com/use-cases/), [Comparisons](https://www.memtro.com/compare/), [Changelog](https://www.memtro.com/changelog/)
- [MCP server](https://www.memtro.com/mcp/) (one MCP server per connector, 750 pages), [AI gateway](https://www.memtro.com/gateway/), [Agent memory for companies](https://www.memtro.com/memory/), [Glossary](https://www.memtro.com/glossary/), [Features](https://www.memtro.com/features/), [Industries](https://www.memtro.com/industries/)
- [Pricing](https://www.memtro.com/pricing/), [Security](https://www.memtro.com/security/), [Privacy](https://www.memtro.com/privacy/), [Subprocessors](https://www.memtro.com/subprocessors/), [About](https://www.memtro.com/about/), [Contact](https://www.memtro.com/contact/)
- [Blog](https://www.memtro.com/blog/): [Code mode: fewer steps, smaller bills, better answers](https://www.memtro.com/blog/code-mode-read-path/), [Why AI hallucinates about your own company data](https://www.memtro.com/blog/ai-hallucination-company-data/), [What the EU AI Act asks of your data](https://www.memtro.com/blog/eu-ai-act-record-keeping/), [MCP vs API, when the caller is a model](https://www.memtro.com/blog/mcp-vs-api/), [Prompt caching, and why your cache never hits](https://www.memtro.com/blog/prompt-caching/), [Your AI pilot did not stall because of the model](https://www.memtro.com/blog/why-ai-pilots-stall/), [What it costs to paste your company into every prompt](https://www.memtro.com/blog/cost-of-pasting-context/), [A knowledge graph is not a better vector search](https://www.memtro.com/blog/knowledge-graph-vs-rag/), [When your AI gets it wrong, can you prove what it knew?](https://www.memtro.com/blog/what-did-the-ai-know/)