bablsoft/accessflow Quick start Get AccessFlow
Compare

An open-source alternative to Teleport Database Access: reviewing the statement instead of the session

Teleport's Database Access connects engineers to databases with short-lived certificates instead of passwords, records the session, and lets them request roles or resources for a limited time. AccessFlow does something adjacent but different: it decides whether a particular statement may run, and under what masking and row filters, before it runs. Both can be self-hosted and both publish their source, but the licences are far apart. This page compares the two on capability and on licence, and says where Teleport is the better choice.

Get AccessFlow All comparisons
How to read this page. Product names are the property of their respective owners, and AccessFlow (made by bablsoft) is not affiliated with or endorsed by any of them. Everything said about another product comes from that vendor's public documentation as it read on the date in Sources; where we could not find a capability documented, the table says Not documented rather than No.

The short version

In one line: Teleport governs the connection with certificates and recordings; AccessFlow governs the statement with review, masking and a chained audit log. Their licences differ as much as their designs.

Choose Teleport when you already run it, or intend to, for SSH, Kubernetes and applications, and you want databases behind the same door: one identity, one set of roles, one audit stream, and engineers connecting from DBeaver, pgAdmin, DataGrip or psql through a local proxy with certificates that expire on their own. Teleport's role model can restrict which database accounts and which databases a person may use and, with automatic user provisioning, which tables they may touch; access requests let them ask for more for a limited time, with two reviewers required if you want; and every session is recorded, with per-statement events for PostgreSQL that can be replayed. Choose AccessFlow when the control you need is on the statement rather than on the connection: every query parsed, checked against written rules, risk-scored by an AI analyzer, routed by policy to a multi-stage approval chain, and executed under column masking, row-level security and export controls, with a hash-chained audit log to prove none of it was altered afterwards.

The licence axis matters here more than on any other page. Teleport's source is published under the AGPL, but the Community Edition binaries are documented as commercial with restrictions, free only for companies under a hundred employees and ten million dollars in revenue, and access requests, single sign-on beyond GitHub, SCIM user provisioning and device trust are Enterprise features. AccessFlow is Apache 2.0, one edition, with every capability on this page included and no threshold on who may use it.

Side by side

CapabilityAccessFlowTeleport
Licence and sourceApache 2.0, one edition, full source, no usage thresholdSource on GitHub under AGPL-3.0; Community Edition binaries are commercial with restrictions, free for companies under 100 employees and US$10M revenue; Enterprise is a paid licence
HostingSelf-hosted only; no hosted editionSelf-hosted (Community and Enterprise) or Teleport Cloud (Enterprise)
How people reach the databaseAccessFlow's web editor, REST API or Model Context Protocol (MCP) server; no wire protocol or driverNative clients and GUI tools through a local proxy started by tsh or Teleport Connect, with short-lived certificates; also the web UI
Engines18 connectors across relational, warehouse, document, key-value, wide-column, search and graph enginesSelf-hosted PostgreSQL, MySQL, SQL Server, Oracle, MongoDB, Redis, Cassandra, ClickHouse, CockroachDB, Elasticsearch and Vitess, plus a large catalogue of AWS, Azure and Google Cloud managed databases, MongoDB Atlas, Snowflake and Oracle Exadata
Every statement reviewed before it runs✓Yes: parse, AI risk score, routing policy, approval, then execution, for reads and writes alike?Not documented: access requests approve a role or a resource for a period; per-statement review before execution is not described
AI risk review of the statement✓Yes, with Anthropic, OpenAI, Ollama, any OpenAI-compatible endpoint or Hugging Face, plus multi-model voting?Not documented
Approval chains and routing✓Multi-stage chains per datasource, plus ordered routing policies that auto-approve, auto-reject, require extra approvals or escalate◐Access requests with configurable approver thresholds and dual authorisation, notified through plugins (Enterprise); ordered multi-stage chains: Not documented
Just-in-time, time-bound access✓Yes: self-service requests, automatic expiry, optional pre-approval of covered queries◐Yes: role and resource access requests with a configurable duration (Enterprise; the Community Edition can request roles from the command line only)
Break-glass with mandatory retro-review✓Yes: per-user, per-datasource, admins included; every admin is paged; a different admin must sign off afterwards?Not documented
Can requesters approve their own request✓Never, on any channel, for any role✓Never: users cannot approve their own access requests
Written SQL review rules✓14 named rules for the relational engines, at off, warn or block per environment; a block forces human review and never rejects?Not documented
Column masking✓Yes: full, partial, hash, email and format-preserving strategies; admins are masked too unless explicitly revealed?Not documented
Row-level security✓Yes: the statement is rewritten with a parameter-bound predicate, failing closed where an engine cannot◐Table-level permissions per role through database object labels and auto-provisioned users; row-level: Not documented
Result export controls✓Yes: allow, watermark, row cap or deny per datasource?Not documented
Audit and security event management (SIEM)✓Every query and decision; entries chained with a keyed hash; append-only writer role; Splunk, syslog, HTTPS and S3 Object Lock sinks◐Session start, query and end events, with the query text recorded; audit export to a SIEM in every edition; tamper-evidence: Not documented
Signed compliance reports✓Yes: PDF and CSV with a detached signature?Not documented
Session recording✕No: there is no interactive session; each statement is its own record✓Yes: database sessions are recorded and PostgreSQL sessions can be played back statement by statement
Hardware keys and device trust✕No◐Yes (Enterprise)
Single sign-on and SCIM✓SAML 2.0, OAuth2/OpenID Connect (OIDC) (Google, GitHub, Microsoft, GitLab, generic OIDC) and SCIM 2.0, all included◐Community: GitHub only. Enterprise: SAML, OIDC, Okta, Entra ID, Google Workspace and others, plus SCIM
Terraform provider✓Yes✓Yes
MCP server for AI agents✓Yes, for every engine, through the same review pipeline◐Yes: an MCP query tool for PostgreSQL databases, tied to a signed-in tsh session
Governed API calls and CI/CD deployment gates◐Yes: outbound REST, SOAP and GraphQL calls (gRPC targets can be registered and reviewed but not yet executed), and a fail-closed deployment gate for pipelines◐Application access exists; per-call approval and deployment gates: Not documented
Schema migrations and GitOps✕No?Not documented
SSH, Kubernetes or desktop access✕No✓Yes: servers, Kubernetes clusters, applications and desktops in every edition

Included With caveats or in a paid edition Not available Not documented

Not documented means we could not find the capability in Teleport's public documentation on the date in Sources; it does not mean the product lacks it. Edition labels follow Teleport's published feature matrix.

Where each one sits in the path to the database

Teleport is an access plane built around certificates. A Teleport agent running the Database Service proxies connections to the databases it protects; an engineer signs in once, and tsh or the Teleport Connect desktop app starts a local proxy that their GUI client or command-line tool connects to, authenticated by a certificate that expires within hours, so no database password ever reaches the laptop. Which database instances, which database accounts and which database names a person may use come from their Teleport roles, and with automatic user provisioning Teleport can create the database user on the fly and grant table-level permissions from labels on imported objects. When someone needs more, an access request asks for a role or a specific resource for a set time, one or two reviewers approve it through whichever chat or ticketing plugin is configured, and the elevated permissions lapse on their own. Throughout, the session is recorded: start, each query, and end, with PostgreSQL sessions replayable statement by statement, and the whole event stream exportable to a SIEM.

AccessFlow does not proxy the connection; it proxies the statement. Whatever arrives, through its editor, its REST API or its MCP server, is parsed first, so that the statement type and the exact tables and columns it references are known before anything else happens. The statement is checked against the datasource's allow-list and, on the relational engines, against fourteen written review rules, then sent to the AI analyzer for a risk score and a list of issues, then handed to a routing policy that decides whether the request is auto-approved, auto-rejected, escalated or sent to a review plan for people to decide, with named stages and a minimum number of approvers per stage. When it runs, row-level security rewrites the predicate, masking rewrites the result, and export rules decide what may be downloaded. Every step is an audit entry chained to the previous one with a keyed hash and written by a database role that cannot update or delete, so an investigator can verify the chain rather than trust it.

The gap on AccessFlow's side is connectivity: there is no local proxy, no certificate flow and no way for pgAdmin to connect through it, and the roadmap lists a native wire-protocol gateway as unscheduled work. The gap on Teleport's side, as far as its documentation goes, is everything that happens to the statement itself: no pre-execution review of a query, no AI opinion on it, no masking of what comes back, and no rewriting of the rows it may see.

Where AccessFlow is the better fit

  • The decision is about the statement. A support engineer may be allowed to connect and still not be allowed to run this UPDATE without a second person; AccessFlow reviews the query, Teleport reviews the access.
  • You want an AI opinion and written rules before a person spends time. A risk score, the issues found, a suggested rewrite, and a catalogue of named checks that can force human review without ever rejecting on their own.
  • What comes back needs shaping. Column masking with no admin bypass, row-level security rewriting the query, and export governance that can watermark, cap or deny a download; none of these is documented for Teleport's database access.
  • Auditors ask whether the log could have been edited. Keyed hash chaining, an append-only writer role, a verify endpoint, signed PDF and CSV reports, and a write-once S3 sink with signed segments.
  • AI agents should be governed like people, across every engine. AccessFlow's MCP server covers all eighteen engines and puts an agent's statements through the same review and audit; Teleport's MCP database tool is documented for PostgreSQL only.
  • Everything must be included, at any company size. Access requests, SAML and OIDC sign-in, SCIM and the rest are in AccessFlow's single Apache 2.0 edition; in Teleport they are Enterprise, and the free Community binaries carry an employee and revenue threshold.

Where Teleport is the better fit

  • You already run Teleport for SSH, Kubernetes or applications. Adding databases to the same plane gives one identity, one role model and one audit stream; AccessFlow would be a second system with its own login and its own log.
  • Passwordless, native-client access. Engineers keep DBeaver, pgAdmin, DataGrip, Compass or psql, and no database credential ever lives on a laptop; AccessFlow has no native-client path at all.
  • Session recording and replay. A recorded session that can be played back statement by statement is the evidence some compliance regimes expect, and AccessFlow has no interactive sessions to record.
  • Cloud-managed databases with cloud identity. Teleport's catalogue of RDS, Aurora, Redshift, DynamoDB, ElastiCache, Cloud SQL, AlloyDB, Spanner and Azure databases, with IAM-based authentication and automatic user provisioning, is far broader than AccessFlow's eighteen connectors.
  • Hardware keys, device trust and per-session MFA. If posture checks on the connecting device are part of the requirement, Teleport documents them and AccessFlow does not have them.

Licence, editions and what is gated

Teleport's feature matrix describes three editions. The Community Edition's code is on GitHub under the AGPL-3.0, but the matrix describes the binaries as commercial with restrictions: free to use for companies with fewer than a hundred employees and less than ten million US dollars in annual revenue. In that edition, single sign-on is GitHub only, access requests are limited to requesting roles from the command line, and SCIM, hardware keys, device trust and moderated sessions are absent; database, Kubernetes, SSH and application access, session recording and audit export to a SIEM are included. Teleport Enterprise, self-hosted or as Teleport Cloud, adds SAML and OIDC providers, the full access-request workflow with its notification plugins, SCIM provisioning, device trust and hardware-key support, and support contracts.

AccessFlow is released under the Apache 2.0 licence, which is permissive rather than copyleft, in one edition, with no employee, revenue or seat threshold. The approval chains, SAML and OIDC sign-in, SCIM, masking, row-level security, audit sinks and signed reports on this page are all in the same download. The trade-off is that there is no vendor behind it: support is the public issue tracker, and the project is much younger and narrower than Teleport. What is available now and what is planned →

Running both

These two fit together more naturally than they compete. Teleport can remain the way engineers reach infrastructure, including the database sessions they need native tools for, with certificates, recordings and access requests governing the connection. AccessFlow can be where a statement goes when it needs a written decision rather than a session: the query a reviewer must sign off, the change an on-call engineer wants approved at night, the request an AI agent submits over MCP, and the API call or the deployment that should not proceed on one person's word. Both export to the same SIEM, so the recorded session and the approved, hash-chained statement sit side by side. If you can only have one, ask whether your risk is in who can connect or in what runs once they have.

Sources

Checked on . AccessFlow's own claims come from its source repository and the pages of this site they link to. If a vendor's documentation has changed since, please open an issue and we will correct the page.