Mercury Agent
A soul-driven, terminal-first AI agent that runs a permission-gated toolset, a local SQLite Second Brain, and a fleet of persistent Mercury
Mercury Agent earns its place if you've been burned by agents that announce "task complete" without proof — the completion contract, shell blocklist, and per-action prompts are the trust layer most rivals skip. The trade-off is real configuration work: YAML, a CLI-first workflow, and a self-hosted daemon. If you want a managed coding agent with zero setup, Claude Code will get you there faster; if you want a fleet of bots you actually govern, this is the more honest tool.
Verified 4d ago · liveness 76/100 · cite: rightaichoice.com/tools/mercury-agent
- Developers automating multi-step coding workflows who want per-action approval before writes, pushes, and shell commands
- Power users who start a task in the terminal and check on it from Telegram, Discord, Slack, or Signal
- Teams running a fleet of persistent Mercury Bots for release notes, fact-checking, and daily briefs
- CI/CD and ops engineers needing a scriptable, always-on daemon with cron schedules and crash recovery
- Users who want a managed, zero-config coding agent — Claude Code gets you working in minutes with no YAML
- Teams needing a documented air-gapped or fully offline deployment they can certify
- Beginners uncomfortable with terminal commands, YAML config, and reading install scripts
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Mercury Agent if you want a no-code, visual agent builder or you need a fully offline air-gapped deployment — it requires terminal comfort and YAML config to get real value.
Shared memory pool, remote dashboard access, and auto-rotating JWTs sit behind Mercury Cloud, so single-machine local use is the only free path.
Mercury Agent is MIT-licensed open source, so the core agent carries no licence fee; your real cost is the model provider spend on your own keys plus any Mercury Cloud subscription for shared memory and remote dashboard access. That makes it cheaper at the tooling layer than commercial agent platforms but potentially more expensive in provider tokens if you run heavy multi-agent fleets with high concurrency.
In short
Mercury Agent — A soul-driven, terminal-first AI agent that runs a permission-gated toolset, a local SQLite Second Brain, and a fleet of persistent Mercury. Best for Developers automating multi-step coding workflows who want per-action approval before writes, pushes, and shell commands, Power users who start a task in the terminal and check on it from Telegram, Discord, Slack, or Signal, Teams running a fleet of persistent Mercury Bots for release notes, fact-checking, and daily briefs. Free to start; paid plans from $20/mo.
What's new in Mercury Agent
Checked 3 days agoAcross the latest 3 updates: 2 feature updates and 1 changelog entry.
v1.2.3 - Completion Contract and Forced-Action Escalation
Every task now ends with a verified verdict backed by a build, typecheck, or test that ran, or an honest pause naming its blocker — eliminating fake 'Task complete' messages.
v1.1.13 - Improved Autopilot and Standalone Installer
Enhanced autopilot detects stuck loops more reliably, and a standalone binary installer was added for macOS, Linux, and Windows.
v1.1.12 - Multi-Agent File Locks and Token Saver Mode
Reader-writer file locks let concurrent agents coordinate without clobbering each other, and Token Saver Mode reduces token consumption during long runs.
What people actually say about Mercury Agent — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
35 mentions across 5 sources (Hacker News, YouTube, Bluesky, GitHub, Lemmy) · researched Jul 14, 2026.
Average across the 5 sources that answered — each source counts once, not each post.
- +Keeps all memory local in SQLite – no cloud dependency.
- +Runs 24/7 as a daemon with multi-channel access.
- +Permission-hardened tools ask before write or shell commands.
- +Token Saver Mode and per-session budgets prevent runaway costs.
- +Flexible permission modes: Ask Me, Allow All, Safe whitelist.
- −Only a few community posts – limited real-world validation.
- −GitHub issue: reasoning_content bug breaks some LLM providers.
- −48 open issues suggest incomplete polish for production use.
- −No Node.js required but still requires technical setup.
- −Learning curve for non-developers unfamiliar with daemon agents.
- • LLM API costs (e.g., OpenAI, Anthropic) – not included in tool pricing.
- • Self-hosting infrastructure (server, electricity, bandwidth).
Viability Score
How well maintained and how widely used is Mercury Agent? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- 25-step agentic loop with graduated escalation to sub-agents
- Parallel sub-agents in isolated context windows
- Reader-writer file locks for concurrent agent coordination (v1.1.12)
- Mercury Bots: persistent specialist agents with their own model and memory (v1.3.0)
- Durable job queue so work survives restarts and can be replayed
- Three permission modes: Ask Me, Allow All, Safe Whitelist
- Shell blocklist refuses sudo, rm -rf /, and flag-swapped variants in every mode
- Folder-level read/write scopes per channel and per sender role
- Second Brain memory: SQLite + FTS5 local store in ~/.mercury
- 10 memory types with auto-extraction, conflict resolution, and hourly consolidation
- Personality in user-owned markdown files: soul.md, persona.md
- Completion contract (v1.2.3): verified by build/typecheck/test, or honest pause naming the blocker
- Forced-action escalation harness to stop model narration
- Token Saver Mode and daily token budgets with live usage
- Provider fallback mid-task across 10+ model providers
About Mercury Agent
Mercury Agent is a self-hosted AI agent built for people who want an autonomous worker without handing it the keys. You drive it from a terminal (`mercury`), a local web dashboard with Kanban and a bot cockpit, or through Telegram, Discord, Slack, and Signal — all six channels share one tool registry, one permission system, and one memory store. Reads are prompt-free; anything that writes, pushes, or shells out waits for an answer. The repo now ships v1.3.0 with Mercury Bots, a fleet of persistent specialist agents that each carry their own persona, model, memory, and permissions, plus a durable job queue so work survives restarts. The pitch is control. Permission modes (Ask Me, Allow All, Safe Whitelist) sit next to a shell blocklist that refuses sudo and rm -rf / in every mode, folder-level read/write scopes per channel and sender role, and bots that fail closed — a missing permission is a denial, not a pop-up. Roughly 50 built-in tools cover filesystem, shell, git, GitHub PRs and issues, Spotify, and Kanban, with 126+ community skills on the open Agent Skills spec installable in a command. Memory is the other half. A Second Brain in SQLite + FTS5 under ~/.mercury extracts preferences, people, and decisions as you work, resolves contradictions, and consolidates hourly; personality sits in markdown files (soul.md, persona.md) you own. A v1.2.3 completion contract means tasks end verified by a build, typecheck, or test that actually ran, or paused with the blocker named. It works with any model from a list that spans Anthropic, OpenAI, DeepSeek, Grok, GitHub Copilot, and local Ollama, with provider fallback mid-task. MIT-licensed and self-hosted, it competes with terminal-native coding agents like Claude Code and Codex CLI on transparency and multi-channel reach rather than on polish and zero-config onboarding.
Behind the Verdict
We'd reach for Mercury when the job is long, multi-step, and touches things we care about — a repo, a PR, a scheduled routine. The permission model is the reason. Reads run free, writes and pushes stop for approval, and the shell blocklist holds even in Allow All mode, so an overeager model can't quietly `rm -rf /`. That's the difference between an agent you leave running overnight and one you babysit. The completion contract (v1.2.3) matters more than any benchmark. A task ends verified by a build, typecheck, or test that ran, or it pauses and names the blocker. In practice, that means fewer silent failures and fewer fake green checkmarks — the exact failure mode that makes most agent output useless in a real pipeline. Mercury Bots in 1.3.0 is the headline for teams. Run a release-notes bot, a fact-checker, and a deploy bot as a fleet, each with its own model and memory, and reply from whichever channel you're on. Start a task in the terminal, check the status card in Telegram. The durable job queue means a crash doesn't lose the work. Where it bites: this is a terminal-first, self-hosted tool. You'll be editing YAML, reading install scripts, and running `mercury up` to get the daemon on boot. If you don't want that, don't buy it. There's no air-gapped or fully offline story documented, so regulated shops should check before committing. The closest alternative is Claude Code. Claude Code is more polished, more managed, and starts working in minutes; Mercury gives you provider flexibility, a fleet of persistent bots, six channels, and local memory you own. Choose Claude Code for a focused coding assistant, Mercury for an agent platform you govern. For privacy-minded developers, the local SQLite Second Brain plus zero agent telemetry is the strongest argument —
Researching Mercury Agent? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Mercury Agent actually fits — and what changes day-one when you adopt it.
Run 'mercury code ~/projects/acme', ask it to migrate the settings page to a new form library, and watch the live plan checklist while it edits files.
Outcome: Mercury runs the typecheck and the settings test suite, then reports 'Task complete · verified' — or pauses and names the failing test instead of claiming success.
Run 'mercury up' to install Mercury as a boot-starting system service, then schedule a daily-brief bot to Telegram and a release-notes bot for draft generation.
Outcome: Bots work through the durable job queue, survive restarts, and surface a 'needs you' notice when the deploy bot requests shell access it was never granted.
Point Mercury Code at GitHub PRs and issues and let parallel sub-agents split review and changelog work while a shared Kanban board tracks state across Discord.
Outcome: Reader-writer file locks keep concurrent agents from clobbering each other, and the hourly Second Brain consolidation carries billing-migration context into tomorrow's session.
Use Cases
- Automate code reviews and PR merges on GitHub.
- Manage Spotify playlists and search music hands-free via Telegram.
- Orchestrate multi-step data processing tasks with parallel sub-agents.
- Schedule recurring system maintenance scripts with custom skills.
- Collaborate with team members on shared Kanban boards via Discord.
- Audit file changes and run shell commands with permission prompts.
- Monitor and control multi-agent workflows from a single CLI or chat interface.
- Run a fleet of persistent specialist bots with a durable job queue.
Limitations
- Mercury Agent is a self-hosted, terminal-first agent that requires local installation via a shell installer or npm (Node.js 22+); the standalone binary routes are macOS, Linux, Windows, and Termux/mobile, with a hosted Mercury Cloud option instead of a fully offline one.
- Memory lives in local SQLite on your disk and the agent runs on your machine with 0 telemetry reported by the vendor.
- It is explicitly permission-gated — reads are free, while anything that writes, runs, pushes, or leaves scope waits for approval, and a shell blocklist stops commands like sudo and rm -rf /.
- It is a developer-oriented tool requiring terminal access and configuration of model providers.
as of 2026-10-07
Verification history
We have re-verified Mercury Agent 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-checked, vendor evidence unchanged
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Mercury Agent's pricing actually pencils out — and where peers do it cheaper.
Mercury Agent is MIT-licensed open source, so the core agent carries no licence fee; your real cost is the model provider spend on your own keys plus any Mercury Cloud subscription for shared memory and remote dashboard access. That makes it cheaper at the tooling layer than commercial agent platforms but potentially more expensive in provider tokens if you run heavy multi-agent fleets with high concurrency.
Setup time & first value
How long it actually takes to get something useful out of Mercury Agent — broken out by persona, not the marketing-page minute.
On macOS or Linux the one-line installer verifies the SHA-256 and drops a binary on your PATH in a couple of minutes — no Node.js needed. npm installs take a similar path if you already run Node.js 20+. Plan on a longer first session for 'mercury doctor', YAML config, and enabling the local dashboard on port 6174; the real time cost is learning the permission model, not the install.
Switching to or from Mercury Agent
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a plain ChatGPT or Claude chat UI: run the installer, then let Mercury's workspace IDE and git panel replace copy-paste iteration on real files.
- →From a lighter CLI coding assistant: import your repo workflow into Mercury Code's AUTO mode and add the completion contract's verified verdicts to your existing test commands.
- →From cron scripts and shell glue: move recurring jobs into Mercury skills and schedule them with Mercury's built-in cron and heartbeat system.
- →From a self-hosted agent with no memory: turn on the Second Brain and let ten memory types plus hourly consolidation replace your manual notes.
- ↗To a fully offline, air-gapped agent runtime: export your soul.md and persona.md files and your SQLite memory, since Mercury has no offline deployment mode.
- ↗To a no-code visual agent builder: expect to rebuild workflows that currently live in YAML config and CLI commands.
- ↗To a managed cloud agent platform: your local SQLite Second Brain contents need migration into the vendor's hosted memory store.
Integrations
Resources & Guides
Tutorials & Learning
YouTube returned 6 videos for “Mercury Agent”, and we withheld 6: 6 did not mention Mercury Agent. We are showing none, because we could not prove any of them are about Mercury Agent.
Official links
Featured Head-to-Head Comparisons
Mercury Agent vs Spider Cloud
Choose Spider Cloud if you need a cheap, high-speed web scraping API for feeding AI models with fresh data — its Rust engine and unblocker make it a no-brainer for RAG pipelines. Choose Mercury Agent if you want an autonomous coding assistant that runs across chat platforms and requires fine-grained permission control. They serve entirely different problems: one fetches web data, the other automates workflows.
Mercury Agent vs Temporal Ai
If you need mission-critical reliability for long-running AI agents that survive crashes, pick Temporal — its durable execution and human-in-the-loop signals are unmatched. If you want a permission-controlled autonomous assistant that runs 24/7 from chat or CLI for coding and task automation, Mercury Agent's 40+ hardened tools and multi-channel access make it the better choice. Choose based on whether you prioritize fault tolerance or autonomy with guardrails.
Mercury Agent vs Presto Voice
If you run a QSR chain and need to automate drive-thru ordering with proven upsell lifts, Presto Voice is your pick. If you're a developer seeking a permission-controlled autonomous agent for multi-step workflows across CLI and chat, Mercury Agent delivers. They solve completely different problems—choose based on whether you're optimizing a restaurant or automating code tasks.
Popular in AI Governance & Guardrails
Mindgard
Mindgard automates AI red teaming — agent-native reconnaissance, exploit validation, and continuous AI security testing for production AI
Poolside AI
Poolside builds open-weight agentic coding models — Laguna XS 2.1 (33B) and Laguna S 2.1 (118B) — for code that stays inside your security
Olas Network
Co-own autonomous AI agents on-chain: stake OLAS, run agents that trade, hire each other and settle payments.
Frequently Asked Questions
Best-of guides
Used Mercury Agent? Help shape our editorial sentiment research.