Dev & Engineering telegram-bridgegit-worktreessession-resumeprogress-streamingmulti-enginefile-transfervoice-notes

Takopi Telegram Coding Bridge

Run, monitor, and resume multiple coding-agent sessions from Telegram.

FollowAgents review · FARS-2.1
Use with care
62/ 100 5-point scale 3.1 / 5
1 2 3 4 5 6
1Trust12 / 29 · 2.1/5

The README discloses the main Telegram, external-agent CLI, progress-streaming, file-transfer, and worktree channels, while release workflows use scoped GitHub permissions and PyPI trusted publishing; authorship and license attribution are clear. Deductions apply because repository, branch, and file operations are powerful yet no runtime least-privilege model, hazardous-action confirmation policy, token storage/redaction policy, retention policy, or complete recovery procedure is shown. Most dependencies have lower bounds rather than exact pins, and no vulnerability scanning or dependency-update policy is evidenced. Unknown publisher identity was not treated as suspicious.

2Reliability8 / 14 · 2.9/5

The README, package metadata, entry points, CI, and release workflow agree on the product purpose, supported engines, Python requirement, and versioning, justifying full self-consistency credit. Dependencies and external engine prerequisites are identified, frozen synchronization is used in CI, and releases run tests. Credit is reduced because Python 3.14+ and separately installed PATH engines narrow availability, with no documented fallback. No runtime implementation or user-facing error examples are supplied, so failure_messages receives zero.

3Adaptability14 / 18 · 3.9/5

The assistant, workspace, and handoff workflows are concrete, and the repository describes multi-project, worktree, group-topic, and terminal-handoff scenarios thoroughly. Engine prefixes, project commands, branch mentions, and plugin entry points provide reasonably precise selection and extension mechanisms. Deductions reflect missing evidence for ambiguous or conflicting triggers, invalid targets, group authorization boundaries, and platform-specific behavior; environment coverage is largely limited to uv, Python 3.14+, PATH engines, and an OS-independent classifier.

4Convention14 / 18 · 3.9/5

The README has clear feature, requirement, installation, setup, usage, plugin, and development sections with pointers to deeper documentation; package, CLI, version, and entry-point naming are consistent. Installation guidance and the complete MIT license justify full marks for those criteria. Deductions apply because the supplied material contains no FAQ, consolidated limitations, support policy, or standalone changelog. Generated release notes and tag/version validation provide only partial version-history evidence. The named author, issue tracker, and release workflow establish an update path, but no maintainer roster, response commitment, or governance policy is shown.

5Effectiveness9 / 13 · 3.5/5

The product offers a clear convenience layer over several agent CLIs through Telegram, with progress streaming, resume support, queues, cancellation, worktrees, and file exchange. These features plausibly create useful mobile-oriented output and marginal value. Scores are reduced because the evidence consists mainly of feature descriptions and test helpers: it provides no representative rendered outputs, evidence of readability for complex results, performance information, or systematic accounting of subscription, Telegram trust-boundary, Python, and engine-installation costs.

6Verifiability5 / 8 · 3.1/5

Identity, version, dependencies, entry points, license, CI, and publishing claims are traceable to specific supplied files. Engine claims are corroborated between the README and package entry points, while CI and pytest configuration independently show quality gates. Deductions apply because most runtime feature claims lack accompanying implementation files or focused tests in the supplied evidence; the tests shown are only fixtures and event factories. Feature prose also rarely distinguishes supported conditions, exceptions, and intended behavior explicitly.

Evidence confidence: Low Reviewed Aug 25, 2026 Reviewed revision e0123d4f9b25
Before you use it
  • Remote Telegram messages can cause agents to operate on repositories, branches, and files. Do not grant access to sensitive repositories until chat authorization, hazardous-action confirmation, and directory isolation are verified.
  • Bot tokens and existing OpenAI or Anthropic subscriptions involve sensitive credentials; the supplied material does not describe local storage, log redaction, rotation, or incident response.
  • Operation requires Python 3.14+, uv, and at least one separately installed agent CLI. Confirm actual availability in the target environment before deployment.
  • Project dependencies are not exactly pinned in pyproject, and no vulnerability scan is shown. Review the lockfile, transitive dependencies, and plugin provenance.
  • Worktrees and cancellation controls are not transactional rollback. File changes, Git operations, and remote effects may require separate backup and recovery measures.
Review evidence [1][2][3][4][5][6][7][8]
See the full review method →

What does this agent do, and when should you use it?

Takopi is a local bridge between Telegram and Codex, Claude Code, OpenCode, or Pi. It runs the selected engine against projects and Git worktrees while streaming commands, tool activity, file changes, and elapsed time into chat. Sessions can continue inside Telegram or be handed back to a terminal with a resume line. Projects, branches, group topics, and agent sessions can be mapped separately, supporting concurrent work across repositories and worktrees. Deployment is through a local CLI and requires Python 3.14+, uv, a Telegram bot token, and at least one supported engine on PATH.

When takopi starts, its setup wizard connects a Telegram chat using a bot token created through @BotFather, then configures an assistant, workspace, or handoff workflow and a default engine. Messages prefixed with /codex, /claude, /opencode, or /pi select an engine and launch its session in the current or registered project. takopi init happy-gadgets registers a project, /happy-gadgets targets it, and @feat/memory-box places a run in a dedicated Git worktree for that branch. During execution, Takopi streams commands, tools, file changes, and elapsed time, while maintaining a per-agent-session queue; Codex sessions also expose steering and cancellation controls. It accepts Telegram voice notes and scheduled messages, transfers files into repositories or returns files and directories, and resumes work through replies or terminal resume lines.

  1. A developer away from their workstation can dispatch work from Telegram and watch commands, tool activity, and file changes as the local agent runs.
  2. An engineer maintaining several repositories can register each project and target it from any chat location with a /project-name command.
  3. A contributor working on concurrent branches can mention @branch-name so each task runs in a dedicated Git worktree.
  4. A user moving between phone and terminal can continue a thread in chat or copy its resume line into the terminal.
  5. A team using Telegram forum groups can bind separate topics to repository and branch contexts.
  6. A developer using more than one of Codex, Claude Code, OpenCode, and Pi can select an engine per message while retaining a default.

What are this agent's strengths and limitations?

Pros
  • One Telegram interface directly supports Codex, Claude Code, OpenCode, and Pi, with engine selection available per message.
  • Projects, branches, and Git worktrees are first-class execution contexts for concurrent, isolated repository work.
  • It streams commands, tool activity, file changes, and elapsed time instead of exposing only a final response.
  • Stateless resume supports both continued chat interaction and handoff back to a terminal.
  • Telegram voice notes, scheduled messages, group topics, and bidirectional file or directory transfer are built into the documented workflow.
Limitations
  • Python 3.14+ and uv are mandatory, which is a higher runtime requirement than tools supporting commonly deployed Python versions.
  • A supported engine must already be installed and available on PATH; Takopi does not supply the underlying coding agent.
  • Operation depends on a Telegram bot token and chat connectivity, making it unsuitable where Telegram or external network access is prohibited.
  • The source says existing Anthropic and OpenAI subscriptions work but does not document costs, quotas, or subscription compatibility limits.
  • The supplied material does not document permission isolation, hardened authentication, sandboxing, or recovery behavior after engine-command failures.

How do you install or deploy this agent?

Install uv with curl -LsSf https://astral.sh/uv/install.sh | sh, then install Python 3.14 using uv python install 3.14. Ensure at least one of codex, claude, opencode, or pi is on PATH, and run uv tool install -U takopi. Start takopi for the first time and follow the wizard to create a Telegram bot token through @BotFather, choose the assistant, workspace, or handoff workflow, connect a chat, and select the default engine.

How do you use this agent?

Start it inside a repository with cd ~/dev/happy-gadgets followed by takopi. Send the bot a message; prefix it with /codex, /claude, /opencode, or /pi to select an engine, and reply to continue the thread. Register the repository with takopi init happy-gadgets, then target it with /happy-gadgets hard reset the timeline. Use /happy-gadgets @feat/memory-box freeze artifacts forever to run against that branch in a dedicated worktree. Inspect or change configuration with takopi config list, takopi config get, and takopi config set.

FAQ

Does Takopi include its own coding model or agent engine?
No. It is a Telegram bridge and requires at least one of codex, claude, opencode, or pi to be installed on PATH.
Do I need an OpenAI or Anthropic API key?
The source does not require API keys. It says existing Anthropic and OpenAI subscriptions work, but does not specify the authentication process or cost boundaries.
What local access does it need?
It runs agents inside projects and Git worktrees, reports file changes, and transfers files or directories to and from repositories, so it needs corresponding filesystem and command-execution access.
Can it handle multiple tasks at once?
Yes. It supports parallel runs across agent sessions and maintains a queue per agent session; Codex sessions additionally have steering and cancellation controls.
What is the minimum runtime setup?
You need uv, Python 3.14+, a Telegram bot token, and at least one supported engine available on PATH.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents