The useful
fine print.

Becki is a memory layer for your work. It captures decisions, dead ends, commitments and open loops from authorized project ingestion and AI conversations. Astir can also capture them from meetings. Supported, connected AI clients can query the same memory instead of asking you to re-explain the project.

The memory belongs to your account. Becki does not automatically read every AI app or expose every conversation to every model.

Becki runs as an MCP server. MCP is the protocol a compatible client uses to call tools. When an agent calls Becki for context, it can retrieve relevant decisions, prior failures, commitments and open questions. After meaningful work, it can save the new reasoning for later.

Connecting the server is only the first step. Skills or agent rules help the client retrieve before acting and capture afterwards. The setup guide includes the protocol and a verification question.

Each tool generally keeps its own context. What you teach Claude is not necessarily what ChatGPT knows, and what you decide in Codex does not automatically become Cursor's history.

Becki supplies shared context that your connected, authorized tools can ask for. That includes the rejected approach: a path ruled out in one client can be retrieved in another, rather than resurfacing as a fresh suggestion. The benefit depends on the agent actually using those memory tools.

Becki keeps failed and rejected approaches as memory an agent can retrieve. Decisions include the reasoning behind them. MCP lets supported clients use that context without tying the memory to one client.

When a choice comes up again, you can look at the original problem, alternatives and constraints. The entries give you a record to check rather than a model's guess about the past.

No. Becki can connect directly to supported MCP clients. Astir is an agent workspace with conversation, files, terminal, previews and Git review. It can use Becki's memory, but your existing tools can use it too.

Core is the memory-layer plan. Astir is the desktop workspace plan. Team and Enterprise are coming soon.

Capture the new choice with its rationale and resolve the superseded entry. Becki's review process can also propose likely supersessions for you to approve or reject. It does not automatically declare an old decision obsolete.

Approval ranks the old entry below the new one while retaining readable history. Rejection leaves both on equal footing. Unreviewed proposals wait; opening a review is not required for ordinary memory use.

The published guide covers Claude Code, Claude Desktop, Cursor, Codex CLI / IDE / desktop, Zed, VS Code, Raycast and Antigravity. Local setup uses Core's generated configuration. Astir manages its own providers, tools and MCP connections in Connections.

For web clients, the guide describes claude.ai, ChatGPT with developer mode, and Perplexity with a paid custom-connector plan. Each client needs its own authorized connection to the remote MCP server. Verify connector availability in your actual session.

Follow the client-specific setup. Connecting tools inside Astir does not automatically configure every external client.

The supported distinction here is Grok models through a configured xAI provider in Astir. That is different from adding a Becki MCP connector to Grok.com. This site does not claim that a Grok.com connector has been verified.

OpenAI and Claude likewise have two routes: models inside Astir, or supported external clients connected to Becki. Account access and provider credentials are part of the setup.

No. A client must support the integration, be configured, authenticate to your account and use the memory tools. A personal vault and intentionally shared team projects also have different scopes.

A local MCP server is not reachable from every remote environment. The published guide specifically notes that Codex's cloud sandbox cannot reach a local server. ChatGPT web has a separate remote connector route.

Core's memory daemon runs on Windows, macOS and Linux. The current Astir desktop installer is for Mac with Apple silicon. Check the requirements of the installer you download; availability of an agent's CLI is not the same as availability of a packaged desktop app.

Remote MCP lets supported web clients reach your authenticated vault without running the desktop app on that device. That is memory access, not native meeting capture.

Optionally, through Astir, with the permissions and participant consent required for the recording sources you choose. Joining a calendar event does not start recording. Core does not include meeting capture.

Local transcription processes audio on your device; configured cloud transcription sends it to the selected provider. Selected text and synced memory have separate processing paths.

A background process that watches supported AI-tool session logs and extracts decisions, dead ends, commitments and open loops. It can capture context even if an agent did not call the ingest tool.

It is opt-in. Review which work you are authorizing for ingestion in the surface you use before enabling it. Connecting an AI tool alone does not authorize reading all its session logs.

The Mac app keeps local files, recordings and local meeting / commitment stores. The published material also describes a retrieval-optimized cloud copy of vault content, scoped to your Becki account, for connected clients and remote MCP.

Do not read “local” as “nothing ever reaches a server.” Synced vault text that passes the security controls is readable by the service. Selected content can reach configured AI providers for processing. Becki is not zero-knowledge. See Trust and Privacy for the full data path.

Becki does not sell your data or use it to train models. When you connect an AI provider to Astir, that provider's terms and your account settings govern the data you send.

No-training is not zero retention. Review Privacy and Subprocessors for Becki-managed processing; your connected tools have their own terms.

The source FAQ describes ODSR—On-Device Secret Redaction—as a credential screen on vault writes before storage or embedding. Recognized provider keys, JWTs, private keys, labeled Social Security numbers and payment-card numbers are replaced with a typed marker containing the kind and last four. Unknown-shape, high-entropy tokens are flagged rather than automatically redacted.

The published controls include an ingest-path screen, a database trigger and a nightly decoy-secret canary. These are defense in depth, not permission to intentionally place credentials in memory. Never authorize storing secrets or paste keys into a work-context prompt. Read the detailed security model.

No. ODIT (On-Device Identity Tokenization) is separate from content-based credential detection. It applies to covered memory paths where enabled; it does not remove every name from arbitrary text, prompts, captures or audio.

Applicable writes may be held in a sealed quarantine lane pending the required controls. This is not universal end-to-end encryption or proof that plaintext never reached the service. Device approval belongs in your real signed-in account. See Security.

Yes. The source FAQ describes local Markdown memory you can copy, archive or version. Settings → Export provides a full JSON export. You can delete individual entries or the account from Settings, and the corresponding entries are removed from the cloud retrieval index.

For written deletion confirmation, contact support@becki.io. Full retention and deletion details belong in the privacy policy; do not assume subscription cancellation is the same operation as immediate account deletion.

The published product and pricing policy keeps the vault for 60 days after cancellation in case you return, then purges it. Export from Settings before you need the copy. Account deletion and explicit forget / delete actions are separate from that cancellation grace period.

No. Bring-your-own-key concerns the provider used for processing. Embeddings and the rest of the vault pipeline are a separate path. Core includes backend LLM calls in its price.

Astir uses the provider accounts and credentials you configure. Confirm the model and processing support you need with support@becki.io. Choosing a local model does not make every connected tool or memory operation local.

Core is $15/mo. Astir is $25/mo or $250/yr, with a 14-day free trial. Core has no published annual price. Team and Enterprise are coming soon, not available subscriptions. The Team price preview is $39/seat/mo or $390/seat/yr for 5–49 seats; Enterprise is custom.

See Pricing for the plan comparison, cancellation policy and availability notes.

Core provides shared memory in your editor and chat tools without a desktop workspace or meeting capture. Astir brings an agent workspace, configured models and tools, files, terminal output, previews and Git review together, with Becki's memory available.

You can use the same Becki account and vault in Astir. You do not need to rebuild the memory.

Team and Enterprise are coming soon. The planned offering includes shared team memory, selected project sharing, invitations and seat administration. These are not available subscriptions yet. See planned Team and Enterprise capabilities or contact support@becki.io.

Use Sign up for Core or Astir. The existing subscription route is unchanged. Use the download portal for the current Astir installer; Core's install token comes from the signed-in Account page. Then follow the client-specific setup guide.

For account questions or to request removal, email support@becki.io. For connection and product support, use support@becki.io.