agentboards.org
Compare/Devika vs Terragon

DevikavsTerragon

Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.

Devika
Stition AI · Autonomous SWE
#250OSS
Panel
3.6
4 spec wins
Reliability
3.2
Usefulness
3.5
Cost
6.3
Longevity
1.3

“Nineteen thousand stars and nothing since 2025, which is the open-source version of a standing ovation on the way out.”

Terragon
Terragon Labs · Autonomous SWE
#270OSSMCP
Panel
3.7
6 spec wins
Reliability
3.3
Usefulness
4.7
Cost
5.3
Longevity
1.5

“It streamed a live browser view so you could watch the agent work, which was lovely right up until February.”

Spec by spec

SpecDevikaTerragon
Architecture
CategoryAutonomous SWEAutonomous SWE
Runslocalcloud, sandbox
Platformsmacos, linux, windowsweb, macos, linux
Context windownot documentednot documented
Protocols
MCP clientNoYes
MCP serverNoYes
Capabilities
Runs terminal commandsYesYes
Multi-file editsYesYes
Git operationsNoYes
Browser controlYesNo
Sandboxed executionNoYes
Multi-agent orchestrationNoYes
Headless / CI modeNoYes
Models
BackboneClaude, GPT, Gemini, Mistralvia managed agents (Claude Code, OpenAI Codex, Amp, Gemini, OpenCode)
Bring your own modelYesYes
Local modelsYesNo
Cost
Pricing modelbyokmixed
Starts atn/a$25/mo
Free tierYesNo
Bring your own keyYesYesUsers could power agents with their existing Claude or ChatGPT subscriptions, or configure their own API keys.
Openness
Open sourceYesYes
LicenseMITApache-2.0
GitHub stars19,556259

Which one would each critic pick

CriticDevikaTerragonPick
El Juez——not enough reviews
El Amigo3.53.3Devika — Do not adopt: this has been quiet since 2025 and the honest replacement for anyone who wanted an open autonomous engineer is OpenHands.
El Crítico3.53.3Devika — The authors described it as early and experimental from the start, and nothing since has changed that label, which makes it the rare project whose warning aged accurately.
El Profesor3.55.0Terragon — The design routed all agent output through pull requests, so verification borrowed the organisation's existing review process rather than inventing one.
La Inversora3.53.0Devika — Stition AI built the most-forked answer to a funded competitor and never converted any of that attention into a company, which is the recurring story of that wave.
La Jefa3.02.8Devika — There is no supplier, no support and no unattended mode, so this is an engineer's side project rather than anything sixty people could be issued.
El Hacker4.55.0Terragon — Apache-2.0 on the way out, `terry` for local handoff and an MCP server in the box, which is a better goodbye than most companies manage.

Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.

Want a third column? The compare tool handles any two agents. Three-way comparisons are on the roadmap once the spec rows are all verified.