HQbird 2027
HQbird 2027
Everything about your Firebird server on one screen. Backups verified by restore. One monitoring center for all servers. Replication and an AI assistant where a person confirms dangerous actions.
- Firebird 5 · 4 · 3 · 2.5
- Windows and Linux x86-64
- 100% compatible with Firebird
- Server requirements: 2+ GB RAM · 2+ cores · 5+ GB disk
- 0 inbound ports to connect to the center
Summary
Four main ideas
No need to open the server console
Full diagnostics of Firebird and the OS. One server or a group of servers in one interface, locally or in HQCenter.
All servers in one place
Database servers need no inbound ports to connect to the center. Agents get new versions of Firebird and HQbird tools from HQCenter themselves and install them, with rollback to the previous version.
Replication on several nodes
Masters, replicas, lag and data checks in one interface. The master moves to another host without data loss, and applications do not change the connection string.
AI assistant under human control
Works with AI chats and autonomous AI agents. Operations are split by danger level. A person confirms every change. Dangerous operations are confirmed only on the server, over a second channel, not in the chat: with the fbmcp ctl approve command or, on Windows, in a tray window.
Advanced server-level features
View all- Light-weight Monitoring (LWM)Performance of all databases on the server every minute, with no extra load. The fblwmon plugin.
- Update conflicts and deadlocksA separate trace catches conflicts during SQL execution.
- Replica reinitialization under loadThe master does not stop. The database copy for the replica is made automatically, with checks of free space and long transactions. In vanilla Firebird this is manual work.
- Switchover and failoverA planned move of the master to another host without data loss. An emergency move by hand or by a rule. A router sends applications to the current master. In vanilla Firebird this is manual work.
- NGram indexFor
WHERE … LIKE '%found%'in Firebird 5 HQbird. - Metrics export to PrometheusFirebird and server metrics in your Grafana.
New in the interface
- All performance on one screenCPU, memory and disks: totals and Firebird separately, including the disks with databases and temporary files. Firebird metrics for all databases at once.
- Complete backup inventoryAll backups on the server, including those made by other programs. Finds lost copies of databases and backups.
Components
- fbagent
- Agent on the Firebird server: backups, maintenance, metrics, traces.
- HQ Monitor
- Web console. One server locally or many servers in the center.
- HQCenter
- Management center for a fleet of servers.
- HQbird Cluster
- Replication: nodes and the cluster manager (RCM).
- HQbird MCP
- MCP server for AI assistants.
What the DBA gets
- A backup verified by restore; straight to S3, with compression.
- A daily server health report; you start an action right from the report.
- Firebird updates with checks and automatic rollback.
- The answer to “what loads the disk, CPU and memory”.
What the developer gets
- The most expensive queries, grouped by plan.
- SQL errors with the application, address and Firebird function.
- Conflicts and deadlocks during SQL execution.
- AI assistant in the IDE: query plan, index advice, migrations.
What the IT manager gets
- One monitoring center for the whole fleet; no inbound ports needed to connect to it.
- Remote settings with delivery status; updates with automatic rollback.
- Replication for Firebird 4 and 5 and HQbird 2.5 and 3.0, with lag control.
- Per-server access rights and an audit log.
In detail
Feature groups
Nine groups: what HQbird does and why it helps. Screenshots are from the HQ Monitor and HQbird RCM user guides. Server and database names are hidden by demo mode.
A backup is useful only if it restores. Turn on test restore in a gbak job, and the agent restores every new backup and writes the result to the job history. If the restore fails, you see it the same day, not on the day of a crash. For nbackup there is a separate restore job.
A backup can go straight to S3, with no intermediate file on the server disk. A separate file_sync job makes copies to S3, FTP and FTPS. Compression is zip, gz or zstd; a zip archive of a gbak backup can be encrypted (AES-256). The source file is deleted only after the archive is verified. Old copies are deleted by retention rules for each nbackup level.
- 1gbak or nbackupon schedule; one database or all databases in the given paths
- 2Test restorea gbak job option; the backup is really restored
- 3Compressionzip, gz, zstd; the source file is deleted after the check
- 4Storagea folder or a stream to S3; file_sync makes copies to S3 and FTP/FTPS
- 5✓ Verifiedresult in the job history and in the health report
Replace cycle: the database is rebuilt from a backup
A large database fragments over time and collects “garbage”. Replace cycle makes a backup, restores it to an intermediate file and replaces the production database. On schedule, with no manual night work.
The production database goes offline safely: the agent puts it into shutdown, all connections close, and only then the file is renamed. Before the replace, the agent checks the size of the restored database. Dangerous gbak switches are not allowed. You can run your own scripts before and after the cycle.
If a power failure or a restart breaks the cycle, the agent rolls it back or finishes it by itself. If a person is needed, the outcome says exactly that: operator needed.
- Group jobsAll databases in the given paths, with or without subfolders, in one job. If a database has its own job of the same kind (gbak or nbackup), the group does not copy it.
- Multi-level nbackupLevels 0…4 by a simple schedule, cron or a plan. The plan shows in the editor calendar. Retention by level or by whole cycles.
- InventoryAll backup files, including those made by other tools. Shows which database has no fresh copy.
- gbak logs in the consoleThe job shows the last log lines and the history of commands it ran. No need to look for the file on the server.
- Restore from S3The archive format is detected automatically, locally and in S3.
- Overload protectionOne heavy operation per database at a time. If the database is busy, gbak waits up to 60 minutes.
Every 10 minutes the agent takes a snapshot of the server state: CPU, memory, disks, OS limits, firebird.log, transactions, locks, queries, backups and replication. The health report shows it as a matrix for 24 hours, 7 or 14 days. A red cell tells you at once where and when there was a problem.
Each finding explains the cause and gives a hint. Next to it is a button that opens the right job or starts it. From “found” to “fixed” is one screen.
- 1Finding“the database has no fresh backup”
- 2Hintwhat to set, with threshold and value
- 3Action button“Add an nbackup job”
- 4Run nowstart from the console and watch the progress; a queued run can be cancelled
- 5✓ Fixedthe next snapshot is green
What loads the server: Firebird or another program
A slow database is often not the fault of Firebird. An antivirus, a backup by another program or a neighbor service takes the disk and CPU. HQ Monitor shows the load by program and service: reads and writes, CPU, memory.
For disks you see read and write latency, IOPS and disk “stalls”. Separately for the disk with databases and the disk with Firebird temporary files.
- Light-weight Monitoring (LWM)Connections, transactions, reads and writes every minute, without a trace.
- TransactionsOIT/OAT/OST gap and long transactions: a signal before the database starts to slow down.
- State from the database headerShutdown, read-only, sweep, nbackup mode. Metric collectors do not connect to a database in shutdown.
- gstat with trends and forecastGrowth of tables and indexes, export to XLSX.
- Linux limitsTasksMax, NPROC, NOFILE, OOM, threads-max. A common cause of “Firebird crashed under load” is visible before the crash.
- Free space and temp filesA warning before the disk is full.
- firebird.log and replication.logWith filters, in one console for all servers. Thread creation errors and deleted lock files are highlighted.
- Config historyVersions of firebird.conf, databases.conf, replication.conf and a diff.
- Prometheus / GrafanaMetrics export to your existing monitoring.
Sweep, index statistics recalculation, gfix checks and your SQL scripts run on schedule. For sweep and SQL scripts you can set a maintenance window. By default, group sweep has a window (00:15–01:15), and SQL scripts have one too (03:00–04:00). Only one heavy operation runs on a database at a time. Sweep, gfix check and index recalculation skip a busy database until the next run.
A Firebird update is now a routine operation too. The agent waits for the update window and for other jobs (gbak, nbackup, scripts) to finish. Then it stops Firebird, replaces the files and checks the engine. The old files stay in the update backup. If the check fails, the agent restores the previous version itself.
- 1Update windowupdates only at allowed times
- 2Waiting for jobsgbak, nbackup and SQL scripts finish
- 3File replaceFirebird is stopped; the old files go to the update backup
- 4CheckFirebird is running and responds
- 5✓ Doneor automatic rollback
A build that was rolled back on this server is blocked. The agent does not install it again until someone removes the block.
- Run nowBackup, restore, sweep or a check with one button, locally or remotely. You see the progress. A queued or waiting run can be cancelled.
- Server performance testA baseline before and after a change of hardware, disk or settings.
- databases.conf editorAlias tree, parameter checks for the engine version, a hint on a typo, a backup before save. From the center: delivery with the agent reply; edits in the form keep the comments.
- Migration from FBDataGuardSettings of the old HQbird are imported automatically.
- Moving from OS-package FirebirdThe installer finds Firebird from apt or dnf and prints the migration commands. Users and aliases move to HQbird.
- Changes without restartNew settings are checked; on error the agent returns to the previous ones.
A developer does not need server access to know which query to optimize first. The performance report is built from a trace and groups queries by plan. You can sort queries by duration, fetches, reads and writes.
An SQL error comes with the application name, IP address and the Firebird function where it happened. For an update conflict or deadlock during SQL execution, you see the statements of both sides. A separate trace catches them. It logs all modifying statements, so it adds its own load.
How the database changes
gstat runs on schedule. The trends report compares snapshots, shows which tables and indexes grow fastest, and makes a forecast. Index statistics find inactive indexes and indexes with poor selectivity. The “Links” tab of the gstat report shows which queries read an index.
Metadata tracking is off by default; you turn it on in the settings. Every 6 hours the agent compares a new schema export with the previous one and records when the schema changed.
- Long transactionsWhich connection holds a transaction and blocks garbage collection.
- Client versionsWhich versions of fbclient and drivers connect to the database.
- Performance report: comparison and history newTen query rankings, procedures and triggers, executions by hour. Comparison of two periods and the history of one query for 30 days.
- Report on request newThe “Build report…” button builds a report for any period up to 24 hours, with filters by database, duration and program.
- Local API (REST)Status, metrics and job runs from your own scripts and CI.
- AI assistant in the IDEThrough HQbird MCP: query plan, index advice, schema compare, migrations right from Cursor or Claude Code.
The agent sends data to the center itself over outbound HTTPS. Database servers need no inbound ports to connect to the center; all traffic goes over mTLS. This also works for customer servers behind NAT and firewalls.
Settings change from the center. The agent fetches the new configuration, checks it and replies. For each change you see the delivery status and the reason for a rejection. Before it replaces the settings, the agent saves a copy of the old ones.
- 1Change in the consolejob or settings form
- 2SENTsaved in the center; the agent has not fetched it yet
- 3TAKENthe agent fetched it and checks it
- 4APPLIEDapplied; or REJECTED with a reason
- Updates from the centerAgents and products update with signature checks and automatic rollback.
- The center updates itselfIn a maintenance window, with signature checks and automatic rollback. IBSurgeon turns it on; the center owner can forbid it.
- Emails or webhook about new problemsAnd a weekly report for the whole fleet. The center administrator turns this feature on.
- The agent protects the serverIf an agent task takes too much memory or CPU, the agent pauses or stops it. It does not touch backups and data sending.
- Users and access rightsPer server; separate rights to change settings, run jobs and run dangerous operations. Audit log.
- 5 interface languagesEnglish, Russian, German, Portuguese (Brazil), Spanish.
The master sends replication journal segments to the replica. The cluster manager (RCM) watches all nodes: lag, queues, errors. One replica can receive databases from several masters. Firebird 4 and 5 and HQbird 2.5 and 3.0 are supported.
A new replica is created with nbackup, without a master stop and under load. Before the start, RCM and the nodes check free space. The master node waits for long transactions to finish.
You can move the master to another host. Switchover does this as planned and without data loss: RCM itself stops writes, waits for the replica and brings the old master back as a replica. Failover is needed when the master host is lost. A router on the application host always sends applications to the current master.
Control and security
RCM knows more than 20 alert types, and each one has instructions. The data check compares the row count of each table on the master and on the replica. You turn on the periodic check in the settings. The Firebird of a master is never restarted automatically; a replica only in allowed cases.
Switchover and failover go step by step. Before the point of no return, an error or a cancel brings the old master back online. Automatic failover is off by default; you turn it on for each service separately, and it needs witness routers.
For a database without a service, the manual way stays: “Promote to master” shows each step in the database row. If a step stops, you can repeat it.
Nodes talk over mTLS with a shared CA. Certificates renew automatically; keys never leave the host.
- Switchover newA planned move of the master with one button in RCM or one command. No data is lost. The old master comes back as a replica by itself.
- Failover newThe master host is lost: the replica takes the master role. By hand, with an explicit consent to data loss, or by the service rule.
- Routers newHAProxy on the application host. After a master change, the connection string stays the same.
- Replica file moveNo reinitialization; rollback on error.
- Pause and resumeAfter a pause, shipping continues from the last segment, without reinit.
- Cluster setup in three steps newTurn on RCM on one host, open the pairing window on the other hosts, link the hosts on the Link hosts page.
- Hung Firebird newThe node connects to the database itself every minute. The port is open but the database does not answer: a serious alert.
- Bulk parameter changeReplication parameters for many nodes, with a preview.
- Table publicationInclude, exclude and control publication for Firebird 4 and 5.
HQbird MCP connects Firebird to an AI assistant: Claude Desktop, Claude Code, Cursor, Codex and other MCP clients. More than 120 tools, Firebird 2.5–5, and all data of the fbagent agent.
The main rule: a person always decides. Reads need no confirmation and run in a read-only transaction. Every change needs confirmation. Dangerous operations are confirmed only on the server, over a second channel, not in the chat: with the fbmcp ctl approve command or, on Windows, in a tray window. The AI cannot confirm them itself.
fbmcp ctl approve. Needs an open maintenance window. A database replace, an exclusive window and dangerous SQL also need a verified backup no older than 24 h.- 1The assistant asks“replace the database from a backup”
- 2fbmcp checks the levellevel 2: the window is open, the backup is verified
- 3A person on the serverapproves with
fbmcp ctl approve - 4Executionwith a preview of what will change
- 5✓ Journal entryhash chain; one command verifies it
- For DBAsSessions, OIT/OAT/OST transactions, gstat, LWM, firebird.log; backups and test restore; sweep, validation; users and rights with protection against SYSDBA lockout; traces from templates.
- For developersQuery plan and index advice with a re-check; query compile check; results as a table right in the chat; schema and data compare; migrations with a rollback plan.
- Ready-made scenariosHealth check, optimization, security audit, DDL review, migration plan, disaster readiness.
- fbagent dataHealth report, LWM, gstat, performance, deadlocks, backups. The assistant runs jobs and changes their settings; it cannot change credentials.
- Cluster newThe assistant builds a replication cluster through RCM: it opens the pairing window, links the hosts, adds databases and checks the replica. It also starts switchover and failover; a person confirms them on the server.
- Remote modeThe assistant on the admin PC, the MCP server on the database server. HTTPS, a separate API key for each person, an allow-list of addresses. Passwords and database paths do not get into the chat.
- Already includedThe HQbird installer installs it, and the agent updates it. On Linux an unsigned build is not installed.
One command installs Firebird 2.5–5 together with the agent, the console and the MCP server. Windows and Linux, x86-64. The installer opens firewall ports for the new Firebird and HQ Monitor: in Windows, firewalld or ufw. Cluster ports open only when you turn on RCM or the pairing window on the server. You can view the installation plan first: the command shows what it will do and changes nothing.
- Several Firebird instances on one serverA new instance installs next to the existing one, on a free port, with its own agent. The installed one does not change.
- Checks before changesThe installation plan and the config check show problems in advance.
- Windows setup wizard in 8 languages newEnglish, Portuguese, German, Russian, Spanish, Polish, Turkish, Chinese.
- Cloud and Docker newA ready image for each Firebird line: Firebird, the agent and HQ Monitor in one container. Backups go directly to S3. More.
- Linux: least privilegeThe agent runs as the firebird user; root runs only verified scripts.
- Passwords and keys are protectedPasswords, tokens and access keys (for example, for S3) entered in the console are encrypted with the agent key. They are hidden in reports and config snapshots.
- Signed self-updateThe agent does not install a build with a wrong signature. On Linux an unsigned build is not installed. Downgrade protection, rollback on a failed update.
- Updates without HQCenter newA server without a center takes updates from the IBSurgeon update server by itself: the agent, HQ Monitor and the MCP server install by themselves, the Firebird package waits for your command.
- Rights in the security DBWhich users exist and who can create databases; private security DBs are found through databases.conf. You can create a new user right in HQ Monitor.
- Closed networksHQ Monitor works without internet.
- Demo modeHides server names, database names and paths, so you can take screenshots safely. All screenshots on this page were taken this way.
- The server owner is in controlRemote commands can be turned off on the server itself. The same for sending the names of other programs and services.
- Agent request logEach request that changes the server, and each refusal, is one line in a log on the server. Secrets and setting values are not written. Kept for 30 days.
HQbird works on a cloud server the same way as on your own server. For Docker, there is a ready image for each Firebird line: Firebird, the agent and HQ Monitor in one container.
The data is on volumes and stays when you replace the container or the image. The agent connects to HQCenter itself, so a server in the cloud needs no inbound ports for the connection to the center.
- 4 imagesFirebird 5, 4, 3 and 2.5. The images are public: you do not need to log in to the registry.
- First start with one command
docker compose up -d: the container installs HQbird itself. - Several instances on one hostEach one has its own ports and volumes. One instance without load uses about 200–250 MB of memory.
- UpdatesFrom HQCenter or with a new image. Both ways keep the data.
- Cluster in a containerThe instance joins a cluster the same way as a server.
- Backups to S3The backup goes directly to the bucket, with no file on the disk. More.
Good to know
- Automatic failover in the cluster is off by default. You turn it on for each service separately, and it needs routers. A manual failover always needs explicit consent to data loss.
- HQ Monitor has no alerts of its own and no Telegram. The center sends emails and webhooks about problems; the center administrator turns this feature on.
- MCP: a dangerous operation is confirmed on the server with
fbmcp ctl approve. The HQbird installer does not install the tray window. - MCP: the result table in the chat works only in clients that support MCP Apps.